Docs for Agents & LLMs
Everything here is machine-readable, point your agent at it
These docs are built to be read by agents as much as by humans. If you’re integrating MeetStream with Claude, Cursor, or any LLM-driven workflow, start here.
The machine-readable surface
Two MCP servers (use both)
- Docs-search MCP -
https://docs.meetstream.ai/_mcp/server- searches these docs. Read-only, no key needed. - Action MCP -
https://mcp.meetstream.ai/mcp(hosted: sign in with OAuth or send your key) - actually does things: creates bots, pulls transcripts and status, sends messages into live meetings. This is the one that lets an agent debug a real bot session.
In Claude Code, add the action MCP with one command, then the agent skills plugin for MeetStream-specific coding knowledge. The plugin ships skills only, so install it alongside the MCP server:
For a copy-paste guide that points a coding agent at all of this and then builds a full app, see Agent Quickstarts.
Agent-friendly CLI
Every @meetstream/cli command supports --json, so agents can drive the full API from a shell without parsing prose:
Ground rules worth telling your agent
- Auth is
Authorization: Token <key>-Token, notBearer. The word matters. create_botrequiresmeeting_link.bot_nameis optional (defaultMeetStream).- Webhook envelopes carry both
eventandbot_event; every ending arrives once asevent: "bot.stopped"andbot_eventgives the reason (bot_statuscan’t tell a kick from a clean exit).bot.doneis the final event on every path. Details in Debugging Bots. - An
Idempotency-Keyreplay returns 507 with the original bot - treat as success, never retry-loop on it. - Media links are presigned and expire - re-fetch the endpoint, don’t cache URLs.
