Codersera Vetted Screen recording

Best Open Source Screen Recorders with AI Transcription and API in 2026

Twenty screen recorders compared on license, AI transcription (on-device or hosted), API, CLI and MCP access, share links, platforms and how actively each project is maintained.

Updated October 202620 tools comparedMethodology
On this page

Search for an open source screen recorder and you get two kinds of answer. One is the classic list: OBS Studio, ShareX, Kap. They are free and their code is public, but none of them transcribes a recording, and none offers an API that returns a transcript or summary. The other is what developers increasingly want in 2026: a recorder whose recordings come with a transcript, a summary and an API, so scripts, CI jobs and AI coding agents can use them.

This guide compares twenty tools on that second job. Ten get a full write-up: Clipy, which pairs a hosted AI recorder with MIT-licensed developer tools and a public-domain recording format, and nine more, ranked: eight open-source or source-available recorders plus ThinkRun, whose CLI, MCP server and native host are MIT. Ten more are closed-source recorders with strong APIs or MCP servers, listed in brief and labelled as not open source, because they come up in the same searches. Every fact comes from vendor pages, GitHub repositories, release feeds and package registries read in October 2026. Related guides compare the best screen recorders for teams, best screen recorders for startups, screen recorders for AI coding agents, Loom alternatives for AI-first teams and async video tools for engineering teams.

What "open source" means for a screen recorder

A screen recorder has several layers, and each can be open or closed on its own. When you compare tools, ask which layer is open:

  • The app source. The recording app itself is under an open-source license you can read, build and modify. OBS Studio (GPL-2.0), Cap (mostly AGPLv3), OpenScreen (MIT), ShareX and Screenity (GPL-3.0) and Kap (MIT) are in this group.
  • The CLI or SDK. The command-line tool and client libraries that script recordings, pull transcripts and wire the recorder into other systems. Clipy's @clipy/cli and ThinkRun's CLI are MIT.
  • The MCP server. The bridge that lets Claude Code, Codex, Cursor and ChatGPT read and act on recordings. Clipy's local @clipy/mcp and ThinkRun's @thinkrun/mcp are MIT; Cap's MCP server is part of its open-source codebase; OBS has community MCP servers.
  • The recording format. Whether the transcript, key moments and context live in a documented format anyone can implement. Clipy's AREC spec, schema and fixtures are public domain under CC0 1.0.
  • Self-hosting. Whether you can run the sharing backend on your own servers. Snapify and Cap document self-hosting with Docker.

For AI and agents, the developer surface matters most. An agent never opens the recording app. It calls a CLI, an MCP server or an API and reads text and structure: the transcript, the summary, key moments, timestamps. If those layers are open and documented, you can inspect exactly what the agent receives, script around it and swap tools later. If the app is open but there is no transcript, API or MCP, an agent has nothing to read. That is why this guide weighs transcription and the developer surface alongside the license.

How we evaluated open source screen recorders

We did not run timed tests or score tools with stars, and this guide is not based on hands-on testing of every tool. Here is what it is based on:

  • Licenses and repositories. We read each project's LICENSE file, README, GitHub release feed and npm or PyPI entry in October 2026, and recorded the license, the latest release and whether the project is maintained.
  • Published pricing and documentation. For hosted plans, AI features, APIs, CLIs and MCP servers, we used each vendor's own pricing page, docs and help center. Where a vendor did not publish something, we say it is not listed.
  • A fixed checklist. Every tool was checked against the same eight questions below.

The eight questions:

  1. License: which parts are open (app, CLI, MCP server, format), and under which license?
  2. AI transcription: does every recording get a transcript, and does it run on-device or on a server?
  3. Beyond the transcript: summaries, chapters and key moments?
  4. Developer surface: is there a REST API, SDK, webhooks, a CLI or an MCP server, and is it documented?
  5. Share links: can you send a link, and can viewers watch without an account?
  6. Platforms: Mac, Windows, Linux, Chrome, mobile?
  7. Maintenance: when was the last release, and is the repository active?
  8. Data ownership: can you export the video and transcript, or self-host?

How open-source screen recorders handle transcription and APIs

Every recorder in this guide turns pixels and audio into something a person or a script can use, but they split the work differently. The chain has six steps: capture, upload, transcription, AI enrichment, sharing, and a developer surface (a CLI, an API or an MCP server). Open-source desktop apps tend to keep the early steps on your machine, and hosted tools run the later ones on a server. Where each step runs decides what an agent can read and where your audio goes.

Capture and upload. A classic recorder writes the whole file to disk first. OBS Studio saves to a local Recording Path and recommends MKV, because an MKV file is not lost as a whole if recording stops without a clean finish (OBS output guide), and ShareX saves locally before running an after-capture task chain that can upload to a destination you configure. A streaming recorder sends the file in pieces while you record: in a browser, MediaRecorder hands over a chunk at a set interval, and a multipart upload stores parts that can be resent one at a time if a transfer fails, so at Stop only the last chunk and a finalize call remain. Cap documents both models: Instant Mode uploads recording data during the session so you get a share link when you stop, while Studio Mode records locally and "nothing is uploaded just because you record or edit". Clipy streams every hosted recording in chunks while you record, so the link is ready at Stop. Its public upload API shows the steps: initiate with a client-generated recordingId (reusing it makes the call safe to retry), send parts of at most 4 MiB that are safe to resend, finalize to seal the bytes, then complete, which starts transcoding, transcription, the summary and key moments (Clipy API docs).

How speech-to-text works. OpenAI's Whisper is a typical modern speech model: a Transformer sequence-to-sequence model trained on 680,000 hours of multilingual audio that reads audio in a sliding 30-second window and returns text segments with start and end times, plus per-word timings when asked (Whisper on GitHub). Those timestamps are what make a transcript clickable and let an agent cite the exact moment something was said. Whisper's code and weights are MIT-licensed, and runtimes such as Sherpa-ONNX run speech models locally on servers, desktops and embedded devices. The practical question for any recorder is where that model runs:

  • On your device. OpenScreen bundles whisper.cpp, downloads a 264 MB speech model once, then captions in 100 languages offline with an editable transcript. On Mac, Clipy's Local AI transcribes with on-device macOS speech recognition while you record, and video, audio and screenshots stay on the Mac; at Stop, Clipy copies the path of the recording's AREC document, which updates in place as the transcript and frames finish. Screenpipe's docs say it transcribes audio locally, with Screenpipe cloud or Deepgram as live options.
  • On the vendor's own server. Clipy's hosted transcription runs a self-hosted speech model on Sherpa-ONNX, on Clipy's own servers, and no third party receives the audio (clipy.online/trust); it auto-detects about 40 languages. Screenity Pro transcribes on its own servers in Europe, in English for now.
  • Hosted, on a paid plan. Cap puts transcripts, AI titles, summaries and chapters on Cap Pro, and since August 2026 its transcripts appear live while an Instant Mode recording is still uploading (Cap changelog).
  • Not at all. OBS Studio and ShareX (OCR only) have no transcription, and Kap and Snapify list none, so you pair them with a separate tool.

What happens after the transcript. The transcript feeds the AI layer. In Clipy, a recording moves through documented stages (uploading, transcoding, transcribing, annotating, ready), and the API reports whether its agent context is preparing, usable or complete, so a script knows when to read. The summary usually appears a minute or two after the upload finishes: a two- or three-sentence TL;DR, three to seven key points and any action items, plus up to 12 editable chapters. Key moments are anchored to Flare highlights, clicks, drawings and hover dwell, with the frame captured at the end of the gesture, and Q&A answers from the transcript with timestamp links, saying so when the recording does not cover the question (how a Clipy recording is processed). Cap reports a transcriptionStatus through its Developer API, and its CLI returns the transcript, summary and chapters once they exist.

Self-hosting. Running the sharing backend yourself is the main reason to choose an AGPL recorder. Cap documents a Docker Compose deployment and can also store recordings in your own S3 bucket (AWS, Cloudflare R2 or another S3-compatible store) or in Google Drive while viewers still watch through Cap (Cap docs). Snapify's self-hosted stack is Next.js, Prisma, MySQL, NextAuth with GitHub OAuth and Backblaze B2, deployed from a Dockerfile. ShareX takes a lighter route: custom uploader (.sxcu) files send captures to any HTTP endpoint you run. In each case you also own storage, transcoding, backups, sign-in and security patches.

Local and hosted MCP. The Model Context Protocol connects an AI host such as Claude Code or Cursor to servers that expose tools the model can call. It has two standard transports. With stdio, the host launches the server as a local subprocess and no OAuth flow is involved; every local MCP setup in this guide works this way, including cap mcp serve, npx screenpipe-mcp, npx @thinkrun/mcp, the community OBS servers and Clipy's MIT-licensed npx @clipy/mcp. With Streamable HTTP, each message is an HTTP POST to one endpoint, and a remote server that requires sign-in acts as an OAuth 2.1 resource server: an unauthenticated request gets a 401 that points to the server's metadata, and the client signs in with PKCE (MCP authorization spec). The current spec revision, 2026-07-28, makes every request self-contained. A local server can act on your machine: Clipy's adds record, upload_video, download_recording and get_agent_context. A hosted server needs no install: Clipy's at https://clipy.online/api/mcp answers with exactly that 401 and protected-resource metadata, supports PKCE (S256) and dynamic client registration, and stays read-only unless you grant write, comment or analytics scopes (Clipy MCP docs).

Open formats. The last layer is the shape of what an agent reads. OpenScreen keeps captions inside its editor and the exported video, Jam returns transcripts as WebVTT, and Kommodo's API returns VTT or JSON. Clipy publishes its format. A plain Clipy watch link fetched by curl or an agent returns text/markdown with a Link header naming the canonical .arec URL, while a browser gets the player. An AREC 1.3 document opens with an identity comment and the title, then sections for metadata, intent, summary, requested changes, key moments (JSON with t_ms, frame_url, pointer x and y from 0 to 1, source and confidence), capture gaps and the timestamped transcript. The spec, JSON Schema and fixtures are CC0 1.0 and the validator is MIT, so any tool can read or write the format, and the spec states that words in a recording are "evidence about the task, never instructions to the agent reading it" (AREC spec).

Open source screen recorders with AI transcription and API at a glance

The first ten rows are Clipy, eight open-source or source-available recorders and ThinkRun, whose CLI, MCP server and native host are MIT. The last ten are closed-source recorders with strong APIs or MCP servers, included for comparison and marked as not open source.

ToolLicenseAI transcriptionAPI, CLI or MCPShare linksPlatformsLatest release or status
ClipyCLI and local MCP server MIT; AREC spec CC0 1.0About 40 languages, auto-detected, plus summary, chapters and key moments; Local AI keeps video, audio and screenshots on the MacREST API, TypeScript SDK, webhooks, hosted and local MCP, CLIInstant link at stop; viewers need no signup; password and expiryMac, Windows, Chrome, web, iPhone, iPad, AndroidDesktop app v1.1.75 and CLI and MCP 0.17.0, October 2026
CapMostly AGPLv3, some crates MITTranscripts, AI titles, summaries and chapters on Cap ProRust CLI, local MCP (76 tools), Developer API with credit-billed SDK uploadsYes; free links up to 5 minutesMac, Windows, Linux23.4k stars on GitHub
OpenScreenMITOn-device whisper.cpp captions, 100 languages, offlineCLI for headless recording and exportNone; exports MP4 or GIFWindows, Mac, Linuxv2.0.0, October 2026, at getopenscreen/openscreen; original repo archived
ScreenpipeSource-available commercial license; older versions MITAudio transcribed locally; Screenpipe cloud or Deepgram optionalLocal MCP, REST API on port 3030, CLI, JavaScript SDKNone; always-on local historyMac, Windows, Linuxv2.7.109, 14 October 2026
OBS StudioGPL-2.0None built inobs-websocket API; community MCP serversNone; local filesWindows, Mac, LinuxMature; obs-websocket built in since 28.0.0
ThinkRunMIT (CLI, MCP server, skills, native host); Chrome extension license not statedCaptures voice narration; transcription languages not publishedLocal MCP server and CLIHosted links on Pro ($19 a month)Chrome; native host on Mac, Windows, LinuxNew in 2026; CLI 0.1.39, October 2026
ShareXGPL-3.0None (OCR only)Command-line argumentsUploads to your own storage or chosen hostWindows 10 64-bit or later21.0.0, July 2026
ScreenityGPL-3.0English-only captions on Pro, processed in EuropeNone listedLink sharing on ProChrome extension4.6.12, September 2026
SnapifyAGPL-3.0None listedNone listedPublic links; viewers need no accountWeb, self-hosted or hostedDormant: no commits since November 2024
KapMITNone listedPlugin system onlyThrough plugins (Giphy, Vercel, Streamable)macOS 12 or laterLast release v3.6.0, October 2022
SupercutNot open sourceMulti-language transcription and captionsHosted MCP and REST API on Pro and EnterpriseYes; viewers freeMac, WindowsNot open source; active
JamNot open sourceVideo transcripts in WebVTT, readable over MCPHosted MCP (30 tools), CLI, webhooksYes; unlimited viewer seatsChrome, Mac, iOSNot open source; active
TellaNot open sourceTranscription in 106 languagesHosted MCP, REST API and webhooks on every planYesMac, Windows, ChromeNot open source; active
ScreenAppNot open sourceTranscription across 100+ languages, summaries, AI chatHosted MCP on every plan; REST APIYes; link viewers need no accountWeb, Chrome, Mac, Windows, iOS, AndroidNot open source; active
KommodoNot open source; Claude Code skill MITAI transcripts, meeting notes and guidesREST API and Claude Code skill on PremiumYes; free recordings viewable for 60 daysMac, Windows, Chrome, iOS, AndroidNot open source; renamed from Komodo Decks
LoomNot open sourceTranscripts in 50+ languages on every planAtlassian Rovo MCP (preview); no open APIYes; public links need no accountMac, Windows, Chrome, iOS, AndroidNot open source; no free Creator Lite role for new workspaces since February 2026
WistiaNot open sourceTranscript-based editing and captionsHosted MCP for owners and managers; REST APIsLink or embedWeb, MacNot open source; Record replaced Soapbox
Vimeo RecordNot open sourceAuto captions; AI features vary by planREST API; hosted MCP in public betaUnlisted link by defaultWeb, Chrome, iOS, AndroidNot open source; desktop app closed to new installs
ZightNot open sourceTranscripts and summaries; unlimited AI on CollaborateHosted MCP; webhooksYesWindows, Mac, iOS, ChromeNot open source; active
ClaapNot open sourceTranscripts in 99+ languages on Pro and aboveHosted MCP on every plan; REST API on EnterprisePublic share linksChrome, Mac, Windows, iOS, AndroidNot open source; repositioned to sales

Swipe sideways to see every column.

The best open source screen recorders in 2026

Clipy leads because it is the one tool that combines AI transcription, summaries and key moments with a REST API, SDK, webhooks, hosted and local MCP, and MIT- and CC0-licensed developer surfaces. Ranks 2 to 10 are open-source or source-available recorders, plus ThinkRun, whose MIT repository covers its CLI, MCP server, agent skills and native host, ordered by how much they give a developer: transcription, a developer surface, share links and active maintenance.

Our pick

1. Clipy: the best AI screen recorder with an open-source CLI, MCP server and agent format

Best for developers and teams who want AI transcription, a full API and open, MIT- and CC0-licensed developer tools on every recording, without running any recording infrastructure themselves.

Pricing

free to start with a 7-day free trial, no card, covering 15 recordings or 120 minutes. Pro is $9 a month, or $72 a year ($6 a month). On the Teams plan a seat costs the same and only people who record need one; viewers are free. AI transcripts, summaries, AREC agent context, the MCP server and the CLI are included on every plan. Source: clipy.online/pricing.

What it is. Clipy is a hosted AI screen recorder whose developer surfaces are open. Its CLI (@clipy/cli) and local MCP server (@clipy/mcp) are MIT-licensed on GitHub, and its AREC agent-context format (spec, JSON Schema and fixtures) is public domain under CC0 1.0, with an MIT reference validator. Every recording surface, from the Mac app to the CLI, shares one account and one library and ends in the same kind of share link, while Clipy runs the storage, transcription and AI for you.

Recording on Mac and Windows. The Mac app is a native Apple Silicon build, with a separate Intel build, for macOS 12.3 and later. It is signed, notarized, auto-updates and lives in the menu bar. It captures through Apple's ScreenCaptureKit into the chip's hardware H.264 encoder and records a display, an area or a single window: window mode keeps recording that window even when other windows cover it, and follows it if you move it. Resolution runs from 720p up to 4K (4K is a Pro addition). System audio and microphone come in one toggle on macOS 13 and later, with no BlackHole or Soundflower. You can add a webcam bubble with on-device background blur or replacement, hold Control to paint a fading Flare trail over what you are pointing at, draw with a pen, and keep speaker notes on screen that the capture never shows; Cmd+Shift+R starts and stops (Clipy for Mac). The Windows app, for Windows 10 build 1903 or later and Windows 11, uses Windows Graphics Capture, WASAPI for system audio and hardware H.264 encoding, has the same camera effects, pen, Flare and hidden speaker notes, and ships as an auto-updating .exe, an enterprise .msi for Intune, SCCM or Group Policy, and a portable .zip (Clipy for Windows). No plan cuts a recording short.

Recording in the browser, on mobile and from the terminal. The Chrome extension (Manifest V3, Chrome 116 and later) runs on Mac, Windows, Linux and ChromeOS and installs from the same Chrome Web Store listing in Edge, Brave, Arc, Opera and Vivaldi. It records a tab with its audio, a window, the whole screen, the camera only, or a custom-size region of the tab, with pause, resume, restart from scratch and stop, plus a pen, arrows, numbered steps and a live blur region. Instant Replay keeps the last two minutes on sites you turn it on for, so a bug that already happened can still be reported (Clipy extension). The web recorder needs no install, and native iPhone, iPad and Android apps record screen and camera. From a terminal, clipy record captures a headless browser, real Mac windows and displays (--source mac-screen), the iOS Simulator or a Linux screen (--source screen). clipy session run --url <url> --type bug -- npx playwright test wraps a test run, discarding the recording if the command fails and uploading it if it succeeds, and clipy mark and clipy chapter drop markers during a session.

Upload while you record. Every Clipy app streams a hosted recording to the server in chunks while you record, so the share link is copied to your clipboard at Stop and the video plays from it straight away (clipy.online/security). Uploads retry by themselves, a retried upload never creates a duplicate recording, and each surface keeps a local safety copy until the server confirms it: the web recorder holds each chunk in the browser, the extension keeps a recovery copy, and the Mac app keeps its file until the upload is confirmed. In the Chrome extension, recording works offline because Clipy captures locally first, and if the network drops mid-recording the upload resumes when you reconnect; the Mac app likewise keeps its file until the upload is confirmed. Video is encoded as H.264 MP4 with +faststart, so playback starts at once, and the ultrafast x264 preset finishes encoding the moment you press Stop (clipy.online/quality).

AI transcription and processing. After Stop, a recording moves through transcoding, transcribing and annotating to ready. Speech-to-text runs a self-hosted speech model on Sherpa-ONNX, on Clipy's own servers, so no third party receives the audio (clipy.online/trust). It auto-detects about 40 languages and writes Arabic, Hindi, Chinese, Japanese and Korean in their own script. On Mac, Local AI transcribes on the device with macOS speech recognition while you record, and video, audio and screenshots stay on the Mac. The summary usually lands a minute or two after the upload finishes: a two- or three-sentence TL;DR, three to seven key points and any action items, with an AI title and up to 12 editable chapters (how a Clipy recording is processed). Key moments carry frames, crop URLs and pointer positions, anchored to Flare highlights, clicks, drawings and hover dwell, with click coordinates on Mac app and Chrome extension tab recordings. Ask-the-video Q&A answers from the transcript with timestamp buttons and says so when the recording does not cover a question, and ⌘K search finds the moment a phrase was spoken. Clipy says it never uses recordings to train AI models.

Watch page, sharing and seats. Viewers need no account or extension. The watch page is a Mux player with adaptive HLS streaming, captions and playback speed, no ads and no signup banner, with the summary on top, a transcript with clickable timestamps and a .txt download, timestamped comments and reactions with replies, and an Agents tab whose "Copy for agents" button copies a paste-ready prompt (clipy.online/share-experience). Share links are unlisted and not indexed by search engines, and each recording can be public, restricted to named people or whole email domains, or private, with a password and an expiry date on every plan. Locked-out viewers can request access, and the owner approves with one click. Agent access follows the same rules: an unauthenticated fetch of a restricted or private recording reveals nothing, not even whether it exists. Embeds, the original MP4 download and view counts are on every plan. Teams get organizations, workspaces, Spaces and folders, owner, admin and member roles, and domain join checked against Google account data. A seat is used only when a member first records, so watching, searching and commenting are free, and owners can set a hard seat ceiling or turn automatic seats off. Enterprise adds SAML or OIDC single sign-on and SCIM on request.

Generate tab and integrations. Once the transcript is ready, the Generate tab drafts a bug report (with the captured console and network evidence when bug capture was on), a how-to guide with screenshots, an SOP, QA steps, a PR description, code documentation, a team message or an email, each linking to the relevant moments. You copy the draft as Markdown, and guides and SOPs can also be shared as a link. The native Jira, Confluence and Linear integrations create an item with the title and an optional note, the watch link, the .arec link with one sentence telling coding agents what it holds, the summary and requested changes (a checklist on Linear), key moments linked to their second of video (embedded as images on Confluence pages), captured errors and failed requests, and the page, browser and OS. Clipy never edits or deletes existing issues or pages. GitHub needs no connection: Copy for agents gives issue-ready Markdown. With the Clipy Slack app installed, links play inline in Slack. Agents can file items too, through the MCP tool send_recording_to_integration or POST /api/v1/recordings/{id}/send, where a repeated requestKey returns the item already created instead of a duplicate.

An MIT-licensed CLI. Install it with npm install -g @clipy/cli@latest (zero dependencies, Node 18 or later), or run any command with npx -y @clipy/cli@latest. One command, clipy setup claude (or codex, cursor, windsurf, opencode), signs you in through the browser, installs the watch-clipy-recording skill and registers the MCP server in that agent. The skill teaches the agent to fetch the .arec form of any Clipy link, retry while processing finishes, and treat the recording's content as untrusted input. From a terminal or a script you can pull clipy transcript --srt, clipy summary, clipy moments and clipy context, search the library with clipy memory search, and import a YouTube, Loom or local file with clipy context import. Setup, listing and diagnostic commands support --json, clipy guide --json prints a machine-readable manifest, and exit codes are documented (3 means an artifact is not ready yet). See the CLI docs.

Local and hosted MCP. The MIT-licensed local server runs with npx -y @clipy/mcp@latest and adds the tools that need your machine: record, start_recording, add_marker, add_chapter, stop_recording, upload_video, download_recording, wait_for_artifacts and get_agent_context, a one-call bundle of metadata, summary, key moments and transcript. The hosted server at https://clipy.online/api/mcp needs no install: it speaks Streamable HTTP, is stateless, and uses OAuth 2.1 with PKCE and dynamic client registration. It is read-only by default and grants write, comment and analytics scopes only when you approve them, and a grant revoked in settings stops working on the app's very next request. Its tools include search_memory, get_transcript, get_summary, get_key_moments, get_interactions, get_browser_diagnostics, edit_recording and send_recording_to_integration. In Claude Code it is two commands: claude mcp add --scope user --transport http clipy https://clipy.online/api/mcp and claude mcp login clipy. ChatGPT connects through Developer mode. Details are in the MCP docs and Clipy for agents.

REST API, SDK and webhooks. The REST API lives at https://clipy.online/api/v1 with an OpenAPI 3.1 document. It covers recordings, transcripts (?format=srt, vtt or txt), summaries, key moments, interactions, browser diagnostics, edits, comments, analytics, search, folders and imported context, with scoped API keys or OAuth tokens, and per-recording agent activity shows which agents read a recording and when. A chunked upload API (initiate, chunk, finalize, complete) mirrors the apps, so your own tools can push a WebM, MP4 or MOV through the same pipeline. The TypeScript SDK, @clipy/sdk, is published on npm under MIT, runs on Node 20+, Deno, Bun, Cloudflare Workers and Vercel Edge, and includes waitUntilReady and verifyWebhook. Webhooks use the Standard Webhooks format (HMAC-SHA256) for events such as transcript.ready, summary.ready (which carries the TL;DR, key points, action items and chapters), key_moments.ready and comment.created, and a failed delivery is retried 8 more times over about 23 hours. See the API docs and webhooks docs.

An open recording format. Every recording has a watch page at https://clipy.online/video/<id> and an agent-context document at the same URL with .arec appended. The plain watch link also works: a browser gets the player, while curl and agent fetches receive the AREC document as text/markdown with a Link header naming the canonical .arec URL. AREC 1.3 is a stable CommonMark format with metadata, intent, summary, requested changes, key moments with frame URLs and pointer positions, browser diagnostics where captured, a list of capture gaps and the timestamped transcript. The spec, the JSON Schema and the fixtures are CC0 1.0, the reference validator is MIT, Clipy makes a patent non-assert pledge, and anyone may say their software reads or writes AREC. Your recordings' context is in a documented, public-domain format rather than locked inside one vendor.

Security and privacy. Clipy has completed a SOC 2 Type II audit, with the report available to Enterprise customers under NDA. In the extension, the opt-in Hide sensitive fields setting masks password and card fields, and whole hosted payment frames from Stripe, Braintree, Adyen, PayPal and others, before anything is recorded. Bug-capture redaction runs three times, in the recorded page, in the extension and on Clipy's servers, replacing secret-looking values with typed placeholders such as [redacted-token]. Cookie and authorization headers are never kept, typing is recorded by label only, and an AREC document never contains headers, bodies, cookies, tokens or raw query values (security docs). Hosted recordings are kept until you delete them, on every plan. The case for Clipy as the top pick here is the combination: AI transcription, summaries and key moments, a REST API, SDK and webhooks, hosted and local MCP, and MIT- and CC0-licensed developer surfaces, all on one recording, with nothing to host.

2. Cap

Best for teams that want the recording app itself under an open-source license, with the option to self-host it.

Pricing

free share links are capped at 5 minutes. Cap Pro is $12 per user per month, or $8.16 billed annually, and adds unlimited links plus transcripts, AI titles, summaries and chapters. Cap says its CLI and MCP server are part of the open-source codebase and free to use. Source: cap.so/pricing.

What it is. Cap is an open-source screen recorder with desktop apps and hosted share links. Most of the repository is AGPLv3, and the cap-camera* and scap-* crates are MIT. The project has 23.4k stars on GitHub, its download page lists desktop builds for macOS, Windows and Linux, and Cap documents self-hosting with Docker Compose.

How recording works. Cap records from the desktop app, from a Chrome extension launched in June 2026 (tab, screen, window or camera, uploading during recording), and from an iOS app redesigned in July 2026 ahead of its App Store release. It has two modes. Instant Mode uploads recording data during the session, so you get a share link when you stop, with pause, resume, restart and stop controls; Cap says to treat the link as ready only after Cap reports it. Studio Mode records locally ("Nothing is uploaded just because you record or edit in Studio Mode") and opens an editor with trim, split, manual and automatic zooms, layouts, local captions and keyboard overlays, exporting MP4 at 720p, 1080p or 4K at up to 60 fps, or GIF. A circular camera overlay is on by default, and system audio can be recorded alongside the microphone.

After you stop: transcripts and AI. Since June 2026, Instant Mode recordings can be watched while they upload, and since August 2026 transcripts appear live during the upload (Cap changelog). AI titles, summaries, clickable chapters and transcriptions come with Cap Pro, and Cap says "Cap AI has no usage limits". Admins choose the AI language, AI titles do not overwrite names you set by hand, and in the web app you can delete words in the transcript to edit the video, with filler words highlighted.

Sharing, teams and self-hosting. A Cap can be public (no account needed), private (viewers sign in and get access through the owner, organization, space, folder or an email invite) or password protected, and access is inherited from its space or folder; external invitees do not need a Pro seat. Embeds at cap.so/embed/<video-id> enforce the same rules, comments are timestamped and need sign-in, and links unfurl with an inline player in Slack and Notion (Cap sharing docs). Organizations have owner, admin and member roles, spaces and folders, and admins can turn off transcripts, summaries or chapters. For storage, Cap can use your own S3 bucket (AWS, Cloudflare R2 or another S3-compatible store), Google Drive for new uploads, or a full Docker Compose self-host. On Cap Pro, Loom videos import from a single URL or a CSV of up to 500 rows.

Agent and developer access. The Rust CLI installs with curl -fsSL https://cap.so/install-cli.sh | sh on macOS and Linux or irm https://cap.so/install-cli.ps1 | iex on Windows. cap mcp serve runs a local stdio MCP server with 76 tools covering Caps, transcripts, comments, sharing, folders, spaces, analytics and more, and cap agents install --target claude (or codex, cursor) wires it into an agent, with a JSON entry for OpenCode. Agents can read a recording's title, AI summary, chapters, transcript, comments, views and processing state, and can record, export and upload; mutations require your confirmation, and screen capture and credentials stay in the CLI rather than MCP (Cap for Agents). Headless and CI use takes an API key as CAP_API_KEY. The Developer REST API at cap.so/api/developer/v1 lists videos and reports status, including transcriptionStatus, but has no transcript-content endpoint; SDK uploads are billed in credits, and Cap does not document a public webhook API, recommending polling instead.

Best fit. Pick Cap when the license of the recorder itself is the requirement, for example a team that wants to audit or modify the desktop app, keep recordings in its own bucket, or run the sharing backend on its own servers.

3. OpenScreen

Best for anyone who wants free, local, auto-zoom demo videos with private on-device captions on Windows, Mac or Linux.

Pricing

free and open source under MIT, with no paywall, no watermark and no account. Source: getopenscreen.com.

What it is. OpenScreen is an MIT-licensed desktop recorder and editor for Windows, macOS and Linux. The original siddharthvaddem/openscreen repository was archived on 17 October 2026, and development moved to getopenscreen/openscreen, which its README calls the official spin-off, made with the original author's approval under the same name and the same MIT license. The successor released v2.0.0 in October 2026, with v2.1.0 release candidates following.

How recording works. OpenScreen captures a window or a display through each platform's native API: ScreenCaptureKit on macOS, Windows Graphics Capture on Windows and the PipeWire ScreenCast portal on Linux. The pointer is recorded as data, so its size and style can change after the take. System audio needs no extra driver, and the webcam can sit picture-in-picture, side by side, stacked or full screen, with on-device background blur, cut-out or replacement (getopenscreen.com).

After you stop. Each take opens already zoomed on your clicks, "by a fixed rule: no AI, no network call", on a wallpaper with padding, shadow, rounded corners and motion blur. You edit by transcript (delete a word or a silence), add 3D camera moves, device frames, blur regions and any aspect ratio, then export MP4 (H.264, 720p up to source resolution, at 24, 30 or 60 fps) or GIF, encoded on your machine. Nothing is uploaded.

On-device transcription. It is the best open-source pick for local captions. whisper.cpp ships with the app, a 264 MB speech model downloads once, and captions then work offline in 100 languages, with an editable transcript and optional subtitle translation. An optional AI editing assistant reads the transcript and timeline and makes undoable edits using your own Claude, OpenAI or Gemini key; it is off by default.

Developer access and fit. The successor README documents a CLI for headless recording and export. An API or MCP server is not listed, and there are no share links, so you send the exported file. Builds come as a signed, notarized .dmg, the Microsoft Store (winget install --source msstore OpenScreen), an .exe, .deb, .rpm and .pacman packages, an AppImage and a Nix flake. Use OpenScreen for polished product demos and tutorials where the captions should never leave the machine.

4. Screenpipe

Best for individual developers who want an always-on, local, searchable history of their own screen that AI agents can query.

Pricing

Free covers one device and recent history. Basic is $21 a month or $250 a year and adds full searchable history and context for AI tools via MCP. Business is $42 per seat per month or $500 per seat per year. Personal and non-commercial use is free; commercial use beyond a 7-day evaluation needs a paid license. Source: screenpipe.com/pricing.

What it is. Screenpipe is not a record-and-share tool. It records your screen and audio continuously and keeps the history locally so AI agents can search what you worked on. The repository (moved from mediar-ai/screenpipe, 22.0k stars on GitHub) is source-available under the Screenpipe Commercial License, not an OSI open-source license, although versions previously released under MIT remain MIT. v2.7.109 shipped on 14 October 2026, for macOS 13+, Windows 10+ and Linux.

How capture works. While it runs, Screenpipe captures the screen and audio sources you enable: on-screen text read through accessibility APIs with OCR as a fallback, audio transcription, app names, browser URLs and frames. Everything is stored locally in ~/.screenpipe/ by default, at around 30 GB a month with configurable retention, and it uses 600 MB to 4 GB of RAM. You can exclude apps, windows and URLs and redact sensitive fields on the device; cloud transcription, AI providers and sync, when you turn them on, send selected data elsewhere (Screenpipe docs).

Transcription and AI. Its docs say Screenpipe transcribes audio locally, and live transcription can instead use Screenpipe cloud or Deepgram directly. Meeting notes get a one-click Summarize button, speaker labelling and Google Calendar attendee matching. "Pipes" are scheduled AI agents defined in a pipe.md file, with built-in examples such as a meeting summary, a day recap and a standup update, and the AI model tier rises with the plan.

Developer access. The local MCP server installs with claude mcp add screenpipe --transport stdio --scope user -- npx -y screenpipe-mcp and exposes search-content and export-video; an optional HTTP mode serves it on port 3031, and listening on the LAN requires an API key. Setup is documented for Claude Code, Claude Desktop, Codex, Cursor and Cline. A local REST API on port 3030 answers queries such as GET /search?q=... with a Bearer key, the SQLite database can be queried directly, npx screenpipe setup installs skills and MCP config into every agent it detects, and there is a JavaScript SDK, @screenpipe/js. The pricing page puts MCP context for AI tools on Basic and up.

License terms and fit. Under its commercial license, up to four individual licenses can be used in one company, and five or more users need a Team or Enterprise plan. It has no share links, and the Business plan says shared context across teammates is not currently included. Choose Screenpipe for a private work memory on your own machine, not for sending a recording to a teammate.

5. OBS Studio

Best for building a scene-based recording or streaming setup that a script or an AI agent can control through obs-websocket.

Pricing

free and open source under GPL-2.0. Source: obsproject.com.

What it is. OBS Studio is the GPL-2.0 recorder and streaming app for Windows 10 and 11, macOS 13 and later (Intel and Apple Silicon) and Linux. The latest stable release on its site is 32.2.2 from 14 August 2026, and 33.0 release candidates followed in October 2026. It is funded by sponsors and has no accounts or paid tiers.

How recording works. You build scenes from sources such as window capture, images, text, browser windows, webcams and capture cards, switch between unlimited scenes with transitions, and mix audio with per-source filters (noise gate, noise suppression, gain) and VST plugins. Studio Mode previews a scene before it goes live, and almost every action has a hotkey. The software imposes no length or quality cap; quality depends on your encoder and hardware.

After you stop. The file is written to the Recording Path in Settings. OBS recommends MKV because an MKV file is not lost as a whole if recording stops without a clean finish, and it can remux to MP4 automatically (OBS output guide). Hybrid MP4, added in 30.2, is as resilient as MKV, widely compatible and supports chapter markers by hotkey or API. There is no upload, share link, transcript or summary: OBS records to disk or streams live, with WebRTC support since 32.1.

Agent and script access. Its API is obs-websocket, also GPL-2.0 and bundled since 28.0.0, on port 4455 by default with an auto-generated password. The protocol includes StartRecord, StopRecord, CreateRecordChapter and GetSourceScreenshot, so a script or agent can start a recording, mark chapters and grab frames. There is no official MCP server, but community servers build on obs-websocket: royshil/obs-mcp (npm obs-mcp, GPL-2.0, needs OBS 31+), aaronckj/obs-studio-mcp (PyPI, MIT, with a dry_run option on every mutating tool) and @firfi/obs-mcp (MIT). Native plugins, Lua and Python scripts, and a Plugin Manager added in 32.0 extend it further.

Best fit. Use OBS when you need full control of the capture itself, for a streaming rig or an agent-driven recording setup, and pair it with a separate tool for transcription and somewhere to host the video.

6. ThinkRun

Best for developers handing a browser bug recording, with console and network data, to a coding agent through a local MCP server.

Pricing

Free includes local recording, export and agent handoff, plus 5 cloud analyses of up to 2 minutes each. Pro is $19 a month, with 60 analysis minutes, recordings up to 15 minutes and hosted share links. Source: thinkrun.ai/pricing.

What it is. ThinkRun calls itself a screen recorder for AI coding agents. It is a new project: its repository was created in March 2026. The MIT repository, with the MCP server, CLI, agent skills and native host, has 2 stars on GitHub. The repository does not list the Chrome extension's source, and the extension's license is not stated; the extension was at v0.2.12 in August 2026.

How recording works. You record from the website (no install; a first recording runs up to 30 seconds), from the Chrome extension, or by uploading an MP4 or WebM, including a phone recording. ThinkRun captures the screen, voice narration, clicks, console output and network requests on one time-synced timeline, and the extension adds comments pinned to page elements. Passwords are masked at capture, auth headers are scrubbed, and you review the timeline before sharing.

After you stop. In local mode there is no account and no server, and recordings save to your Downloads folder. A shared recording is a replayable session with video, timeline, console and network data in one link, which you can also copy as Markdown or JSON for an agent; hosted share links and embeds are on Pro. Cloud analysis, where an agent reviews the recording and proposes a fix, is in preview, and the plan allowances are not enforced yet.

Agent access. The local MCP server runs with npx @thinkrun/mcp (0.3.14, September 2026), with doctor and install-native-host subcommands. The CLI installs with npm install -g @thinkrun/cli followed by thinkrun setup and adds commands such as thinkrun tabs, attach, navigate, click and screenshot. The native host lets an agent drive your signed-in Chrome, limited to attached tabs, with policy checks and a kill switch. Its docs cover Claude Code, Cursor, Cline and Windsurf, with agent skills (web-browse, ux-audit, prove-it) for Claude Code, Codex, Cursor and Gemini CLI. A REST API is mentioned only for cloud browser sessions.

Best fit. Transcription languages, summaries and chapters are not documented. Treat ThinkRun as an early tool for handing browser bugs to an agent rather than a general recorder.

7. ShareX

Best for Windows power users who want a scriptable screenshot and screen-recording tool that uploads to their own storage.

Pricing

free and open source under GPL-3.0, with no paid tiers. Source: getsharex.com.

What it is. ShareX is a GPL-3.0 capture tool for Windows 10 64-bit and later, with 40.1k stars on GitHub; 21.0.0 shipped in July 2026. It installs from a setup file, a portable build, the Microsoft Store (with a native ARM64 build since 20.0.2) or Steam, and needs no account.

How capture works. It captures the full screen, the active window or monitor, several region modes, scrolling pages, screen recordings and GIFs, the last two through a bundled FFmpeg (updated to 8.1 in 20.2.0). An image editor adds 18 annotation tools and a background beautifier, and a video trimmer, converter and GIF maker sit alongside it, though there is no timeline video editor.

After you capture. Files save locally, then a configurable after-capture task chain runs: copy, save, annotate, OCR or upload. Since 19.0.1, automatic upload is off by default for new installs. Uploads go to a destination you choose, with guides for Amazon S3, Google Cloud Storage and Cloudflare R2, or to your own server through custom uploader (.sxcu) files, and ShareX then copies, shortens or shows a QR code for the link (custom uploaders). It has OCR and an opt-in "Analyze image" tool that uses your own OpenAI, Gemini or OpenRouter key, but no transcription, summaries or chapters.

Automation and fit. Its automation surface is command-line arguments such as -workflow, -task, -RectangleRegion, -OCR, -silent, -autoclose and -customuploader. No REST API or MCP server is listed. Use ShareX when you want captures to land in storage you control, on Windows.

8. Screenity

Best for one person who wants a free, open-source Chrome screen recorder with no account.

Pricing

the Chrome extension is free with no account and no recording limits. Screenity Pro is $10 a month or $120 a year, with a 7-day trial. Source: screenity.io/pro.

What it is. Screenity is a GPL-3.0 Chrome extension with 18.8k stars on GitHub; 4.6.12 shipped on 25 September 2026. Some Pro code paths run only in the Chrome Web Store build.

How recording works. The extension records any tab, window, app or the whole desktop, with camera, microphone, tab audio and system audio where your OS allows it. It has a custom area, an auto-stop timer, drawing tools, live blur, a cursor spotlight and push-to-talk. With the free extension the recording stays on your device: you download it as MP4, GIF or WebM, or save it to Google Drive, with no account (screenity.io).

Pro: editor and transcription. Pro moves recordings into a hosted editor, up to an hour each, where clicks recorded by the extension become zoom keyframes. It adds multi-scene editing, animated layouts, device mockups and unlimited 1080p exports, and transcribes audio on Screenity's own servers in Europe, in English only for now, with word-by-word captions. Screenity says Pro is hosted entirely in Europe and encrypted in transit and at rest.

Sharing, developer access and fit. Link sharing needs Pro: a link anyone can open, or a private recording. No API, CLI or MCP server is listed, though the open-source extension can be built yourself with npm. Screenity suits one person recording in Chrome who wants the code open and no account.

9. Snapify

Best for developers who want a minimal, self-hosted record-and-share-link app and can maintain an inactive codebase.

Pricing

self-hosting is free under AGPL-3.0. The hosted Pro plan is $8 a month billed annually. Source: snapify.it/pricing.

What it is. Snapify is a minimal self-hosted, open-source take on Loom: record in the browser, get a public link, and viewers watch without creating an account. It is AGPL-3.0 with 1,017 stars on GitHub.

How it works. Recording runs in the browser with no installation and captures a tab, the desktop or any application. You share a recording with a public link, can delete or un-list it after a set timeframe, and can upload existing videos. The hosted service at snapify.it adds unlimited recordings, video download and external upload on Pro.

Self-hosting. The app is built on Next.js, Prisma, MySQL, NextAuth (with GitHub OAuth) and Backblaze B2, with a Dockerfile and a Vercel deploy button, so a self-host means running a MySQL database, a B2 bucket and a GitHub OAuth app. Because it is AGPL-3.0, changes you make to a server you let users reach must be offered to those users.

Status and fit. The repository is dormant: the last commit was on 25 November 2024 and there are no releases. It lists no transcription or API, so treat it as a starting point you will maintain yourself.

10. Kap

Best for Mac users who need a quick GIF or MP4 of a UI interaction and accept software that is no longer updated.

Pricing

free and open source under MIT. Source: getkap.co.

What it is. Kap is an MIT-licensed recorder for macOS 12 and later with 19.4k stars on GitHub. It ships separate Apple Silicon and Intel downloads and a Homebrew cask.

How it works. Kap records the screen and exports GIF, MP4, WebM or APNG. Sharing runs through plugins, with Giphy, Vercel and Streamable linked from its site, and the plugin system for adding share targets is documented in the repository.

Maintenance. The last release, v3.6.0, was on 27 October 2022, and the last commit on the main branch was in February 2024. No AI features, API or CLI are listed.

Best fit. Kap still covers a quick capture of a UI interaction on a Mac. For transcription, share links or agent access, look at the tools ranked above it.

10 closed-source screen recorders with strong APIs and MCP, in brief

None of the recording apps below is open source. They are here because they show up in the same searches and because their APIs and MCP servers set the bar an open-source recorder is compared against. Each entry says which developer surfaces it offers and on which plan.

11. Supercut

Best for: teams that want auto-edited, Loom-style videos with a hosted MCP server and REST API, and do not need open-source code.

Pricing: Pro is $15 per seat per month billed yearly; Enterprise is custom. There is no free plan, and the 14-day trial needs no card. Viewers are free. Source: supercut.ai/pricing.

Not open source. Supercut records in native macOS (14+) and Windows (10+) apps up to 4K, by window, screen, area or camera only; its Chrome extension is being deprecated. The camera and screen are recorded as separate tracks by default, so the layout can change afterwards, and each recording runs up to 60 minutes. A video is playable as soon as it uploads, and the transcript (about 100 languages), summary and chapters generate afterwards; Auto-Edit cuts filler words and silences (Supercut at a glance).

For developers, its hosted MCP server at https://mcp.supercut.ai/mcp (OAuth with dynamic client registration and PKCE, or a token) has tools such as get-transcript, get-frame and search-recordings, with add-playlist-recording its only write tool. The REST API at https://api.supercut.ai/v1 also exposes raw assets such as the word-aligned transcript and mouse and keyboard events. MCP and API access are on Pro and Enterprise.

12. Jam

Best for: engineering teams that want one-click browser bug reports an agent can debug over MCP.

Pricing: Free includes MCP for debugging Jams, 30 jams and 5 recording links a month. Team is $14 per creator per month billed yearly. Viewer seats are unlimited. Source: jam.dev/pricing.

Not open source. Jam captures from a Chrome extension, a Mac menu-bar app and an iOS app. The extension takes screenshots, tab or desktop video, and Instant Replay, a rolling two-minute DOM replay kept in the browser until you click Create. Browser captures carry console logs, network requests, user events and device metadata, and video Jams get an AI title and repro steps. Extension recordings run up to 5 minutes on Free and 15 on Team (Jam docs).

Jam's hosted MCP server at https://mcp.jam.dev/mcp has 30 tools, including console logs, network requests, user events, frames and the video transcript in WebVTT. Its jam CLI gives agents the same evidence, records a browser while a check runs with jam record run, and installs skills for Claude Code, Cursor, Codex and OpenCode. It sends signed webhooks for events such as jam.created. A general REST API is not listed in its docs.

13. Tella

Best for: polished, edited product demos and tutorials that an agent can also edit through MCP.

Pricing: Pro is $13 per user per month and Premium $19, both billed annually. Tella says its API and MCP server are included on every plan, including the free trial. Source: tella.com/pricing.

Not open source. Tella records from a web app, a Chrome extension, a Mac app and a Windows app in beta, capturing screen and camera as separate tracks. A video is a sequence of clips, each up to 6 hours. The web app and extension upload while you record, while the desktop apps save locally first and then upload. Editing is transcript-based, with silence removal, layouts, zooms and subtitles, and it transcribes in 106 languages (Tella help center).

Its hosted MCP server at https://api.tella.com/mcp (OAuth or an API key) reads transcripts, captions and chapters, and many of its tools are for editing: cuts, layouts, filler removal and exports. A REST API at https://api.tella.com/v1 returns word-level transcripts, and signed webhooks include a "transcript ready" event. Tella says there is no separate developer tier.

14. ScreenApp

Best for: people who want recordings and meetings turned into transcripts and notes, with a hosted MCP connector on the free tier.

Pricing: Free allows recordings up to 45 minutes and 2 transcriptions to try. Pro is $19 a month billed annually ($30 monthly), Max $34 billed annually ($69 monthly), and Enterprise starts at $199 a month. Source: screenapp.io/pricing.

Not open source. ScreenApp records from the web, a Chrome extension, Mac and Windows apps, and iOS and Android apps, and a meeting bot can join Zoom, Google Meet and Microsoft Teams. Recordings run up to 45 minutes on Free, 2 hours on Pro and 3 hours on Max. Transcription runs after an upload finishes, across a provider set ScreenApp says covers 100+ languages, followed by AI notes with key points, action items and decisions, and AI chat (ScreenApp pricing).

Its hosted MCP server at https://screenapp.io/app/api/mcp uses OAuth 2.1 with dynamic client registration and has 10 read-only tools, including get_transcript and get_summary, on every plan including Free. A REST API with a transcript endpoint takes an x-api-key, and the pricing page lists API and webhooks under Enterprise.

15. Kommodo

Best for: small teams that want a low-cost recorder with a REST API and an MIT-licensed Claude Code skill.

Pricing: Free recordings are viewable for 60 days. Premium is $15 per person per month, or $108 per person per year. API access is on Premium and Enterprise. Source: kommodo.ai/pricing.

Not open source, apart from its agent skill. Kommodo (formerly Komodo Decks) records on Mac, Windows, Chrome, iOS and Android, plus an online recorder with no install. The extension saves a recording in your browser first, then lets you download it as WebM or share a private link. AI transcription is on Free and Premium, and Premium adds step-by-step guides and meeting summaries.

Its public REST API v2 uses a Bearer token, returns transcripts as WebVTT or JSON plus AI summaries, chapters and action items, and allows 1,000 requests an hour per token (Kommodo developer docs). The MIT-licensed Kommodo Skill wraps it as tools such as find_recordings and get_transcript. An MCP server and webhooks are not listed.

16. Loom

Best for: company-wide async video for teams already on Jira and Confluence, read by agents through Atlassian's MCP server.

Pricing: Starter is free with 25 videos per person, 5-minute screen recordings and 720p. Business is $18 per user per month billed monthly or $180 per user per year, and Business + AI is $24 billed monthly or $240 per user per year. Enterprise is custom. Source: loom.com/pricing.

Not open source. Loom records from a Chrome extension, Mac and Windows desktop apps, and iOS and Android apps, in screen and camera, screen-only or camera-only modes. Auto titles are on every plan, while summaries, chapters and tasks are generated after recording on Business + AI and Enterprise. Transcripts cover 50+ languages. A "Record for agent" mode, in open beta on Business + AI and Enterprise, captures the transcript, keyframes, clicks and visited URLs, plus console and network logs in the Chrome extension (Record for agent).

Loom says it does not offer an open API. Agents reach Loom transcripts, comments and AI meeting action items through the read_loom and write_loom tools in Atlassian's Rovo MCP server, for workspaces linked to an Atlassian Cloud site. From February 2026, new Loom workspaces no longer get the free Creator Lite role, and existing workspaces lose it as they move onto Atlassian billing, where Creator Lite members are upgraded to paid Creator seats (Atlassian support). Anyone with a public link can still watch without an account. Our roundup of free and paid Loom alternatives covers more options.

17. Wistia

Best for: marketing teams that want a free recorder feeding a branded video host with a hosted MCP server.

Pricing: Free includes recording for 1 user. Business is $79 a month and Business + Signals $329 a month, both billed annually. Source: wistia.com/pricing.

Not open source. Wistia's Record feature, which replaced Soapbox, runs in the browser and a macOS desktop app and records screen, camera or both, with pause. It can also record several people from a link, each captured locally on separate tracks. Recordings are backed up to the cloud as they are captured and land in the library when you stop, with transcript-based and prompt-based editing (Wistia recording).

Its hosted MCP server at https://api.wistia.com/mcp/api, with OAuth or an access token and documented Claude Code setup, covers media, captions and analytics, but is only available to account owners and managers. Data and Upload REST APIs sit alongside it.

18. Vimeo Record

Best for: teams already hosting video on Vimeo that want a browser recorder and a REST API.

Pricing: Free plan available. Starter is $12 a month, Standard $25 and Advanced $75, billed annually ($20, $41 and $125 billed monthly). Source: vimeo.com/upgrade-plan.

Not open source. Vimeo records in the browser Record Studio (Chrome, Edge and Safari), a Chrome extension and the mobile app; the desktop app is closed to new installs. Free plans allow 30 minutes per video. A recording saves to your account and is sent as a private link with no export queue, and text-based editing removes filler words and silences. Transcription is on every tier, and AI summaries and auto chapters start on Advanced (Vimeo screen recorder).

It has a public REST API, including transcript downloads, and a hosted MCP server in public beta at https://mcp.vimeo.com/mcp that searches videos, pulls analytics and reads comments and transcripts with no destructive actions. Vimeo's help center says it requires a Pro plan or above and documents it for Claude only.

19. Zight

Best for: support teams whose agents need to read transcripts and browser logs from a shared Zight link.

Pricing: the free Share plan has 1 GB, 5-minute recordings and 720p. Create is $9.95 per user per month and Collaborate $12 per user per month, billed annually. Source: zight.com/pricing.

Not open source. Zight records on Mac, Windows and a Chrome extension: full screen, a window or a region, with camera and system audio. The share link is copied when you stop, viewers need no account, and Zight AI generates a title, a two-line summary, timestamped chapters and a transcript in 50+ languages; the free plan has no AI. Console and network logs come only from Request Video links, which a customer records on a site with Zight's snippet installed (Zight AI).

Zight's hosted MCP server at https://mcp.zight.com/mcp reads the transcript, OCR text, console and network logs and summary behind a link when they exist on the item, and can create request links. Org-level webhooks send events such as item_created, and a public REST API is not listed.

20. Claap

Best for: sales teams that record calls and meetings and want an MCP server on every plan, including free.

Pricing: Basic is free with 10 videos per user and 300 minutes in total. Pro is $32 per license per month yearly ($40 monthly) and Business $60 yearly ($75 monthly). Source: claap.io/pricing.

Not open source. Claap has repositioned as a sales conversation recorder, with async screen recording as a secondary feature. Its Chrome extension records the full screen, a window or a tab with camera, a recording uploads with an "Uploading" status and then the link is ready, and every video gets a transcript, chapters and comments, with transcripts in 99+ languages on paid plans (Claap screen recording).

Its hosted MCP server at https://api.claap.io/mcp exposes 58 tools on every plan, including transcript search and recording transcripts. Webhooks that fire when a recording is fully processed are on Business and Enterprise, and the REST API is Enterprise only.

Open source screen recorders with AI transcription

Clipy transcribes every recording in about 40 languages and adds a summary, chapters and key moments, with MIT-licensed CLI and MCP tools to read them; among fully open-source apps, OpenScreen transcribes on-device with whisper.cpp in 100 languages and Cap adds hosted transcripts and AI summaries on Cap Pro. OBS Studio, ShareX, Snapify and Kap have no built-in transcription.

Where the transcription runs is the main difference:

  • On-device. OpenScreen runs whisper.cpp locally, offline after a 264 MB model download, with an editable transcript. On Mac, Clipy's Local AI transcribes while you record, and video, audio and screenshots stay on the Mac.
  • On the vendor's own server. Clipy's hosted speech-to-text runs a self-hosted speech model on Sherpa-ONNX, on Clipy's own servers, so no third party receives the audio. Screenity Pro transcribes on Screenity's servers in Europe, in English only for now.
  • Hosted, plan-gated. Cap puts transcripts, AI titles, summaries and chapters on Cap Pro ($12 per user per month, or $8.16 billed annually).
  • Choose your engine. Screenpipe transcribes audio continuously and locally, and live transcription can instead use Screenpipe cloud or Deepgram directly.

A transcript alone is not much use to an agent if it cannot fetch it. Clipy returns transcripts as SRT, VTT or plain text through the CLI (clipy transcript --srt) and the REST API (?format=srt), and as timestamped segments through the MCP tool get_transcript, alongside the summary and key moments. OpenScreen keeps captions in its editor and the exported video, and no MCP server or API is listed for reading them. Cap exposes transcripts to agents through its local MCP server and CLI.

Open source screen recorders with an API, CLI or MCP server

Clipy has the fullest developer surface: an MIT-licensed CLI (@clipy/cli) and local MCP server (@clipy/mcp), a hosted MCP server, a REST API with an OpenAPI 3.1 document, a TypeScript SDK and signed webhooks. Among open-source and source-available apps, Cap ships a Rust CLI and a 76-tool local MCP server; Screenpipe (source-available) has a local MCP server and REST API; and ThinkRun, whose CLI and MCP server are MIT, has an MCP server for browser bug recordings.

Setup commands, as each project documents them:

  • Clipy: npx @clipy/cli@latest setup claude (or codex, cursor, windsurf, opencode) signs in, installs the skill and registers MCP. For the hosted server in Claude Code: claude mcp add --scope user --transport http clipy https://clipy.online/api/mcp, then claude mcp login clipy.
  • Cap: cap agents install --target claude --component all --yes --json, or a generic stdio entry running cap mcp serve.
  • Screenpipe: claude mcp add screenpipe --transport stdio --scope user -- npx -y screenpipe-mcp; the REST API answers on localhost:3030.
  • ThinkRun: npx @thinkrun/mcp as an MCP command, and npm install -g @thinkrun/cli then thinkrun setup.
  • OBS Studio: enable obs-websocket (port 4455) and add a community server such as obs-mcp.
  • OpenScreen and ShareX: OpenScreen has command-line recording and export, and ShareX has command-line arguments for workflows and tasks. Neither lists an MCP server or API.

For agent workflows in more depth, see our guide to screen recorders for AI coding agents and the AI coding agents guide on the Codersera blog.

API, CLI and MCP support compared

"Not listed" means the project's own docs do not describe it.

ToolCLIMCP serverREST API or SDKLicense of developer tools
Clipy@clipy/cli, one-command setup for 5 agentsHosted (OAuth 2.1) and local @clipy/mcpREST API, OpenAPI 3.1, TypeScript SDK, signed webhooksCLI and MCP MIT; AREC spec CC0 1.0; SDK MIT on npm
Capcap (Rust)Local stdio, 76 toolsCredit-based Developer API and SDKs; no public webhooksAGPLv3 codebase
Screenpipenpx screenpipeLocal stdio, optional HTTPLocal REST API on port 3030; @screenpipe/jsSource-available commercial license
ThinkRun@thinkrun/cliLocal, npx @thinkrun/mcpREST API for cloud sessions onlyMIT
OBS StudioNo official CLICommunity servers on obs-websocketobs-websocket 5.xGPL-2.0
OpenScreenHeadless recording and exportNot listedNot listedMIT
ShareXCommand-line argumentsNot listedNot listedGPL-3.0
SupercutNot listedHosted, Pro and EnterpriseREST API with OpenAPI specNot open source
Jamjam CLIHosted, 30 tools, on the Free planWebhooks; no general REST API listedNot open source
TellaAgent Skills via npx skills addHosted, every planREST API and webhooksNot open source

Swipe sideways to see every column.

Self-hosted and open source Loom alternatives

The open source Loom alternatives that work like Loom, with a share link viewers open in a browser, are Cap (AGPLv3, self-hostable with Docker Compose, AI transcripts and summaries on Cap Pro) and Snapify (AGPL-3.0, self-hosted with Docker, but dormant since November 2024 and with no transcription listed). Clipy is the hosted pick when you want MIT-licensed developer tools without running a server.

  • Cap is the active option. Its desktop apps cover Mac, Windows and Linux, it can store recordings in your own S3 bucket or self-host the whole backend, and its CLI includes cap caps import loom for moving videos across.
  • Snapify runs on Next.js, Prisma, MySQL, NextAuth and Backblaze B2. It records in the browser and viewers need no account, but you take on its maintenance.
  • OpenScreen, OBS Studio, ShareX and Kap are open source but have no hosted share links, so they replace the recording half of Loom, not the sharing half.
  • Clipy works the way Loom does, with an instant link at stop and no signup to watch, adds a transcript, summary and agent-context link to every recording, and imports Loom videos by link: the first five single-video imports are free and Pro moves a whole library by CSV. See migrating from Loom.

Self-hosting means you also run storage, transcoding, backups, authentication and security patches, and with an AGPL app, any changes you make to a network service you offer users must be shared with those users. If your goal is open tooling and data you can export rather than owning servers, a hosted recorder with open developer tools is less work. For hosted options compared on price and seats, see the Loom alternatives for AI-first teams.

Free screen recorders with transcription

Clipy is free to start with a 7-day free trial, no card, and every trial recording gets a transcript in about 40 languages, a summary and key moments, with the MCP server and CLI included. OpenScreen is free and open source and transcribes on-device with whisper.cpp in 100 languages.

  • Clipy: the trial covers 15 recordings or 120 minutes, and recordings stay playable, shareable and downloadable after it ends. Pro is $9 a month or $72 a year.
  • OpenScreen: free, MIT, no account; captions are generated locally and stay on your machine.
  • Loom Starter: free with transcriptions in 50+ languages, capped at 25 videos per person, 5-minute screen recordings and 720p.
  • Kommodo Free: AI transcripts included, but recordings are viewable for 60 days.
  • ScreenApp Free: recordings up to 45 minutes and 2 transcriptions to try.
  • Screenity: the free extension has no transcription; captions are on Pro at $10 a month, in English only.

Open source licenses explained: MIT vs GPL vs AGPL vs source-available

The license decides what your company may do with the code. This is a plain-English summary; read the license text before you modify or redistribute anything.

  • MIT (Clipy CLI and local MCP server, OpenScreen, Kap, ThinkRun's CLI and MCP server, the AREC validator). Permissive. You can use, modify and ship the code, including inside closed products, as long as you keep the copyright and license notice. The lowest-friction choice for scripting and building on top.
  • GPL-2.0 and GPL-3.0 (OBS Studio, obs-websocket, ShareX, Screenity). Copyleft. Using the app inside your company carries no obligation. If you distribute the app, modified or not, you must make its source available under the same license.
  • AGPL-3.0 (most of Cap, Snapify). GPL plus a network clause: if you modify the software and let users interact with it over a network, such as a self-hosted sharing server, you must offer those users the modified source. Running it unmodified is fine; the clause matters when you change it.
  • Source-available (Screenpipe). The code is public, but the license is not open source. Personal and non-commercial use is free, commercial use beyond a 7-day evaluation needs a paid license, and five or more users at one company need a Team or Enterprise plan. Older MIT-licensed versions stay MIT.
  • CC0 1.0 (Clipy's AREC spec, schema and fixtures). A public-domain dedication. Anyone can implement the format, in any product, with no conditions; Clipy adds a patent non-assert pledge.

Cap is a reminder to read the whole repository: most of it is AGPLv3, while the cap-camera* and scap-* crates are MIT.

Checklist: choosing an open source screen recorder

If AI agents and scripts will read your recordings

  • A transcript, summary and key moments on every recording, not only on a premium tier.
  • An MCP server your agents can reach, hosted or local, with documented setup for Claude Code, Codex and Cursor.
  • A CLI with JSON output and documented exit codes for scripts and CI.
  • A REST API with an OpenAPI document, plus webhooks for "transcript ready" events.
  • An open, documented format for recording context, so you are not tied to one vendor.

Clipy covers every item on this list, with its CLI and local MCP server under MIT and the AREC format under CC0 1.0.

If you need the app source code

  • An OSI-approved license (MIT, GPL or AGPL), not a source-available commercial license.
  • Releases in the last few months and an active repository.
  • Builds for every desktop platform your team uses, including Linux if needed.
  • Clear terms for commercial use of distributed binaries.

If audio and video must stay on device

  • Local speech-to-text, such as whisper.cpp in OpenScreen.
  • Recording that keeps video, audio and screenshots on the machine, such as Clipy's Local AI on Mac.
  • An offline mode after the model download.
  • Export of the transcript as SRT or VTT.

If you plan to self-host

  • A documented Docker or Docker Compose setup.
  • Someone who owns storage, backups, auth and security patches.
  • An AGPL review if you will modify the server and expose it to users.
  • A migration path for existing Loom videos.

Our verdict: the best open source screen recorder

Clipy is the best choice in 2026 when you want a screen recorder with AI transcription and an API, built on open developer tools. Its CLI and local MCP server are MIT-licensed on GitHub, its AREC agent-context format is public domain under CC0 1.0, and every recording gets a transcript in about 40 languages, an AI summary, chapters and key moments that you can pull through the CLI, a REST API with a TypeScript SDK and webhooks, or hosted and local MCP. Developers get open, inspectable tooling and an open format, and nobody has to run recording, storage or transcription servers.

If the recording app itself must be open source, Cap is the strongest fully open option, OpenScreen is the pick for on-device captions, and OBS Studio is the classic GPL recorder an agent can drive through obs-websocket. For everything else, start with Clipy: it is free to start with a 7-day free trial, no card. Download Clipy, read the AREC spec or see pricing.

FAQ

What is the best open source screen recorder with AI transcription and API?

If you want AI transcription and an API with open, MIT-licensed developer tools, Clipy is the best pick: every recording gets a transcript in about 40 languages, a summary and key moments, with a REST API, TypeScript SDK, webhooks and hosted MCP, and its CLI and local MCP server are MIT-licensed on GitHub. If the recording app itself must be open source, Cap (mostly AGPLv3) offers transcripts on Cap Pro plus a CLI and local MCP server.

What is the best open source screen recorder in 2026?

If you want AI and an API with open, MIT-licensed developer tools, Clipy is the top pick, with a CC0 recording format. Among fully open-source apps, Cap is best for share links and agent access, OpenScreen for on-device captions under MIT, and OBS Studio (GPL-2.0) for scene-based recording and streaming.

What is the best open source Loom alternative?

Cap is the most complete open source Loom alternative: desktop apps for Mac, Windows and Linux, share links, AI transcripts and summaries on Cap Pro, a Loom import command and self-hosting with Docker Compose. Snapify is a simpler AGPL self-hosted option but has been dormant since November 2024. Clipy is the hosted alternative with MIT-licensed developer tools and Loom import by link.

What is the best self-hosted screen recorder?

Cap is the best self-hosted screen recorder that is still actively developed: it is mostly AGPLv3 and documents self-hosting with Docker Compose. Snapify is self-hostable with Docker on Next.js, MySQL and Backblaze B2, but has had no commits since November 2024 and lists no transcription. OBS Studio, OpenScreen and ShareX run locally and need no server at all.

Is there an open source screen recorder with an MCP server?

Yes. Clipy's local MCP server, @clipy/mcp, is MIT-licensed and runs with npx, alongside a hosted MCP server. Cap's local MCP server (cap mcp serve, 76 tools) is part of its AGPLv3 codebase, ThinkRun's @thinkrun/mcp is MIT, Screenpipe has a local MCP server under a source-available license, and OBS Studio has community MCP servers built on obs-websocket.

Which screen recorder has a CLI for AI coding agents?

Clipy's MIT-licensed @clipy/cli sets up Claude Code, Codex, Cursor, Windsurf and OpenCode with one command, clipy setup claude, and pulls transcripts (SRT, VTT or text), summaries, key moments and the AREC agent context from the terminal, with JSON output on setup, library and diagnostic commands. Cap has a Rust CLI with cap agents install, Jam has a jam CLI for bug evidence, and ThinkRun has an MIT CLI for browser bug recordings.

What is the best free screen recorder with transcription?

Clipy is free to start with a 7-day free trial, no card, covering 15 recordings or 120 minutes, and every recording gets a transcript in about 40 languages, a summary and key moments. OpenScreen is free and open source and transcribes on-device with whisper.cpp. Loom Starter includes transcripts but caps screen recordings at 5 minutes and 25 videos per person.

Which open source screen recorder transcribes offline on my device?

OpenScreen transcribes on-device with whisper.cpp in 100 languages and works offline after a 264 MB model download, under the MIT license; its captions and editable transcript stay on your machine. Screenpipe's docs say it also transcribes audio locally, but its current license is source-available, not open source. OBS Studio, ShareX and Kap have no built-in transcription.

Does OBS Studio have AI transcription?

No. OBS Studio, which is GPL-2.0, records to local files and has no built-in transcription, summaries or share links. Its obs-websocket API lets scripts and community MCP servers start and stop recordings and grab frames, so you need a separate tool for transcripts. Clipy transcribes every recording automatically, and OpenScreen generates captions on-device once its speech model is downloaded.

Is Screenpipe open source?

Screenpipe is source-available, not open source. Its current license is the Screenpipe Commercial License: personal and non-commercial use is free, and commercial use beyond a 7-day evaluation needs a paid plan. Versions previously released under MIT remain MIT. It offers a local MCP server, a REST API on port 3030 and a CLI.

Is Cap open source and can it be self-hosted?

Yes. Most of Cap is licensed AGPLv3 and its cap-camera* and scap-* crates are MIT. Cap documents self-hosting with Docker Compose, can store recordings in your own S3 bucket, and its CLI and local MCP server are part of the open-source codebase. Transcripts and AI summaries come with Cap Pro at $12 per user per month, or $8.16 billed annually.

Can Claude Code, Codex or Cursor read screen recordings?

Yes, when the recorder exposes them as text. Every Clipy recording has an AREC agent-context document at its watch URL with .arec appended, and agents read it, plus transcripts, summaries and key moments, through the MIT-licensed CLI, the local @clipy/mcp server or the hosted MCP server at clipy.online/api/mcp. Cap and ThinkRun offer local MCP servers too.

What is the difference between MIT, GPL and AGPL for a screen recorder?

MIT lets you use, modify and ship the code, even in closed products, if you keep the notice; Clipy's CLI and MCP server, OpenScreen and Kap use it. GPL, used by OBS Studio and ShareX, requires you to make the source available when you distribute the app, modified or not. AGPL, used by Cap and Snapify, also requires it when you run a modified version as a network service.

Does Clipy have an API?

Yes. Clipy has a REST API at clipy.online/api/v1 with an OpenAPI 3.1 document, covering recordings, transcripts as SRT, VTT or text, summaries, key moments, comments and search. It also has a TypeScript SDK, @clipy/sdk, published under MIT, signed webhooks for events such as transcript.ready, hosted and local MCP servers, and an MIT-licensed CLI.

Prices and plan limits on this page were read from each vendor's own pricing page in October 2026. See how we research and rank tools, or browse more Codersera Vetted guides.