Connect Claude to your meetings
One key and one endpoint. After this, an assistant can start a meeting, read what was actually said — with the right name on every line, because each microphone is transcribed separately — and answer in the room while it is running.
Make a key
Shown once and stored only as a hash. If you lose it, make another — there is no way to read it back, including for us.
Point your assistant at it
What it can then do
start_meeting · list_meetings · get_transcript
· get_notes · send_message · speak
· draw · list_recordings
Ask it to start a meeting and it hands back a link you can send. Ask what was decided and it reads the real transcript rather than a summary of one. The full protocol, webhooks and the socket API are on the developers page.
Don't drive the meeting UI — use the key
An agent that clicks around the page fights the call for the microphone, and it reads captions off the screen instead of the transcript, which is where the speaker names are actually correct. Everything a computer-use agent would try to do by clicking is a tool call above, and those work while the meeting is running.
The one thing worth doing in a browser: joining a room to be seen. Open
teammeet.io/<code> like a person would. Everything else goes
through the key.
What it cannot do yet
No raw audio and no video — an agent gets text, per speaker, as it is said. It only works in meetings held here; it cannot join someone else's Zoom. That is the trade that makes ground-truth speakers possible at all: we are not a guest in somebody else's room.