composio-google-meet
GitHub指导通过 Composio 调用 Google Meet API 获取会议转录和录音。强调需按空间、记录、转录 ID 的三步链式流程操作,处理分页及异步空结果状态。
Trigger Scenarios
Install
npx skills add memoe-app/memoe --skill composio-google-meet -g -y
SKILL.md
Frontmatter
{
"name": "composio-google-meet",
"author": "Moe",
"license": "MIT",
"version": "1.0.0",
"metadata": {
"hermes": {
"tags": [
"Composio",
"Google Meet",
"Transcripts",
"Recordings",
"MCP"
],
"toolkit": "googlemeet",
"related_skills": [
"composio-router"
]
}
},
"description": "Read before any GOOGLEMEET_* call through Composio. You cannot ask for a meeting's transcript or recording directly — there is a three-step chain through the conference record, and skipping a step is why the obvious call fails. Also: an empty transcript list often means 'not ready yet', not 'none'."
}
Google Meet through Composio
Read composio-router first for the method.
Byte costs on this page are unverified. The schema facts were read from Composio on 2026-09-04 and hold for anyone; sizes need a connected Google Meet account, and this machine had none. Transcript entries are the one thing here that can grow without bound — a long meeting is a lot of segments — so page them rather than asking for all of them.
Moe narrates nothing while this toolkit runs. googlemeet is not one of
the toolkits SpokenStep.place() has words for, so a spoken turn goes quiet
for the whole chain below — and this chain is three or four calls long. Say
what you are doing yourself.
The chain — there is no shortcut to a transcript
A space is the meeting room; a conference record is one occasion that room was used. Everything you actually want hangs off the record, not the room, and there is no tool that takes "yesterday's standup" and returns its transcript.
1. GOOGLEMEET_LIST_CONFERENCE_RECORDS → conferenceRecords/{id}
2a. GOOGLEMEET_GET_TRANSCRIPTS_BY_CONFERENCE_RECORD_ID → transcript ids
2b. GOOGLEMEET_LIST_RECORDINGS (parent: "conferenceRecords/{id}")
3. GOOGLEMEET_LIST_TRANSCRIPT_ENTRIES (conference_record_id + transcript_id)
→ the actual speaker / timestamp / text segments
Step 3 is the only one that returns words anybody said. Steps 1 and 2 return
identifiers. GOOGLEMEET_GET_CONFERENCE_RECORD_BY_NAME fetches one record
when you already have its name; GOOGLEMEET_GET_MEET describes a space,
which is the room and not the meeting.
Note the shape of LIST_RECORDINGS.parent: the full resource name
conferenceRecords/{id}, not the bare id. LIST_TRANSCRIPT_ENTRIES wants
the bare conference_record_id and the transcript_id. Two tools in one
chain, two different conventions.
An empty result is often "not yet"
GET_TRANSCRIPTS_BY_CONFERENCE_RECORD_ID says it plainly: transcripts need
processing time after a meeting ends, and an empty list may be transient.
It may also mean transcription was never switched on, or the organiser's
domain forbids it. Do not report "there is no transcript" from one empty
call on a meeting that just finished — say it is not available yet, and
offer to look again.
Recordings have the same property, and they land in Google Drive as MP4 —
so a recording that exists in Meet is also findable through
composio-google-drive.
GOOGLEMEET_GET_MEET warns that a newly created space returns incomplete
data; retry after a second or two rather than reporting a broken meeting.
Page sizes
Every list tool here takes page_size with no default set by Composio,
so the API's own default applies: 25 for conference records, 10 for
recordings, 10 for transcripts, 10 for transcript entries. All are capped at
100, and values above 100 are coerced rather than refused.
Ten transcript entries is a few sentences of a meeting. Expect to page
through page_token for anything real — and when paginating, every other
argument must match the call that produced the token.
Transcript entries are people talking
They carry names and what was said. They belong in the answer to the person who asked and nowhere else — not a note, not a filename, not a hand-off.
Version History
- e2735d6 Current 2026-11-20 21:42


