Rec over MCP: connect the library, then add the recorder

Rec has two MCP connections with different jobs. The hosted server gives your assistant access to your workspace. The local recorder gives it a browser to drive and record. Understanding that split makes setup much clearer.

The hosted connection: library and sessions

Connect to https://recapp.dev/api/mcp through a client that supports remote MCP. Complete the Rec OAuth sign-in. The hosted server exposes nine tools for recordings, transcripts, summaries, and session management.

Your assistant can use search_recordings to find a walkthrough, then get_transcript or get_summary to read it. Session tools can create a session, add annotations, obtain an upload URL, and finish processing. This connection does not open a browser on your computer.

The local recorder: browser and capture

The @recapp/mcp package runs locally with a Rec API key. It provides browser navigation, snapshots, clicks, typing, and native video capture. Your agent can call record_start, walk through the app, annotate the steps, and call record_stop to upload the video.

Use a client that can run local MCP servers, such as Claude Code, Claude Desktop, or Cursor. The setup examples name this server recapp-recorder, so it can sit alongside the hosted recapp connection.

Recording a product behind a login

The local recorder’s authenticate tool opens a visible browser window. You sign in there; the agent resumes in the authenticated session. The browser profile can reuse that login until the session expires.

Choose the connection for the task

To find a recording or read a transcript, use hosted MCP. To have Rec launch and record the browser, add the local recorder. To do both from one assistant, configure both connections. See the agent setup guide for Claude Code commands, the desktop bundle, and a Cursor configuration example.