speak puts words into a live conversation with colleagues or customers. The guard here is economic — a per-user daily cap — rather than a human approving each utterance, which is worth understanding before wiring it to an autonomous agent. The relationship half is more quietly impressive: personal details carry source attribution, so you can see where the system learned something about someone, which is the minimum for data of that kind.
Groupthink's endpoint, with 28 tools enumerated in their catalogue. Two halves: a bot that actually joins meetings and participates, and a relationship record about the people in them.
- join_meeting and leave_meeting — send the bot to a meeting URL, returning a bot id; leaving clears the transcript cursor
- get_transcript — new lines since the last call, with a per-user cursor
- speak — the bot says something aloud via text to speech, under a per-user daily spend cap
- send_chat — a chat message from the bot, up to 2000 characters
- bot_status, save_notes, get_notes — the bot's state, and idempotent note storage
- Relationships: list, get, search and insights — personal details with source attribution, tags and notes, plus outreach needs
- get_upcoming_meetings — with the attendees' relationship context attached
A bearer API token per the documentation; our probe also found OAuth 2.0 with PKCE and dynamic client registration, scope mcp:use.
Paste a URL, then authorize — add the endpoint to your client, then approve the OAuth consent screen