Model Context Protocol

Give your AI a
screen recorder.

Rec ships an MCP server with 18 tools. Claude, Cursor, and any MCP client can search your video library, read transcripts, and record real browser demos, even behind a login.

Included on every plan. Agents count as $0 teammates.

Two-line setup

Local for Claude Code and Cursor, or hosted with OAuth for everything else.

terminal

# Claude Code

$ claude mcp add recapp --env RECAPP_API_KEY=rk_… -- npx -y @recapp/mcp

 

# Or any MCP client, hosted with OAuth 2.1

https://recapp.dev/api/mcp

What your agent can do

Search & read

Your agent can answer questions from everything your team ever recorded.

  • search_recordings: semantic search across titles & transcripts
  • get_transcript: full transcript with speaker labels
  • get_summary: AI summary with key timestamps
  • list_recordings: browse the workspace library
  • search_sessions: search agent annotations & steps

Record

Your agent drives a real browser and captures native video, not screenshots.

  • record_start: launch a browser, start capturing
  • browser_navigate / click / type: drive the app like a user
  • record_note: annotate each step on the timeline
  • record_stop: upload, summarize, return a share link

Authenticate

Record gated apps like Stripe, HubSpot, or your own dashboard without handing credentials to the AI.

  • authenticate: you sign in once, in a visible window
  • persistent profile: auth is reused for future recordings
  • auth detection: agents detect login walls and ask, never guess

The agent recorder, step by step

You ask your assistant for a demo. It calls record_start and Rec launches a real browser with video capture rolling. The agent navigates your app the way a user would, clicking, typing, and waiting for screens to settle, then drops a record_note on the timeline at every step so the recording explains itself.

If the app needs a login, the agent doesn't ask for your password. It calls authenticate, a visible window opens, you sign in once, and the session is saved to a local profile for next time. The AI never sees your credentials.

On record_stop, Rec uploads the video, writes a summary with key timestamps, and generates a natural AI voiceover from the annotated timeline. What comes back is a share link to a finished, narrated demo, ready for Slack, a changelog, or a customer.

What teams automate with it

Release demos on autopilot

Merge a PR, then tell your agent to record a walkthrough of the new feature and post it to Slack. Every release ships with a narrated video and nobody had to hit record.

Docs that never go stale

Regenerate your onboarding and how-to videos whenever the UI changes. The agent re-records the same flow and Rec adds a fresh AI voiceover, in five languages if you want them.

Competitive teardowns

Point an agent at a competitor's product and get back an annotated, searchable recording of their flow. Your whole team can query it later: “how does their checkout handle taxes?”

A queryable video knowledge base

Ask Claude “what did we decide about the auth flow?” and it answers from your team's recordings, with timestamp citations. No tabs, no scrubbing.

Built like infrastructure, not a toy

OAuth 2.1 with PKCE for hosted clients, scoped API keys with expiration for local ones, and row-level security on every table underneath. Agent logins happen in a window you control, so credentials live in a local browser profile. Never in a prompt, never in our database in plaintext.

Your agents are ready. Give them a recorder.

Sign up free, grab an API key, and your assistant can make its first recording in the next five minutes.

Start free