meeting

name: meeting description: Locate a meeting recording or transcript given a description of it (title, approximate time, attendees). Resolves the calendar event via the Google Calendar MCP, finds the matching entry in four-star, and fetches the transcript Doc via the Google Drive MCP.

four-star (binary: star) indexes Oxide meeting recordings and transcripts from Google Drive. Typical ask: "look at the transcript of my meeting with Peter earlier" or "find the recording of last week's product eng sync."

four-star locates the recording/transcript metadata; full transcript content is pulled from Google Drive via the Drive MCP (four-star itself doesn't expose transcript text).

The CLI and API have several rough edges (see "Known issues" below). Note them in-line when they affect the answer, but do not try to work around them beyond what this skill documents and do not open PRs against ~/oxide/four-star.

Workflow

Resolve the meeting via Calendar MCP — always. Never assume the user's phrasing is a substring of the recording title. Use Google Calendar MCP tools to find the event from whatever the user gave (title keywords, rough time, attendees, topic) and extract the canonical title and UTC start date. This step is mandatory even when the description looks like a title — the user might say "the product meeting" for an event titled "Product Roundtable," or "my meeting with Curtis" for an event with no useful title keyword at all. Without resolving first, title-substring matching against four-star is unreliable.
Find the four-star entry. Run find.sh <YYYY-MM-DD> [title-substring] (next to this SKILL.md), passing a distinctive substring of the resolved title from step 1. It lists recordings in a ±1-day window around the date (widen with --widen N if needed), filters by title substring, and prints a JSON array with name, created_at, recording_id, transcript_external_id, and the drive link. Auth failures surface as four-star's own "Not authenticated" stderr message with a non-zero exit; transcript_external_id is null when not indexed.
Fetch transcript text. four-star itself exposes no endpoint for full transcript retrieval — only cropped transcript_matches snippets via /search. Instead, take the transcript_external_id from step 2 and pass it to the Google Drive MCP read_file_content tool. Transcript docs are Google Docs; the MCP returns speaker-tagged text directly. Access works as long as the user's Google account can open the Doc (typically the case for meetings they attended).

When transcript_external_id is null, the recording's name still pins down the transcript Doc precisely. Meet names paired Docs identically except for the trailing word: replace - Recording with - Transcript and search Drive by that exact title (e.g. title = 'Foo - 2026/04/24 09:58 PDT - Transcript'). The recording's timestamp prefix makes this unique even for recurring meetings, so prefer it over re-deriving a title from the calendar event.

Fallback if no Drive MCP is available: search --q "<distinctive phrase>" returns transcript_matches.blocks[].lines[] with speaker/line fields around the query hit. Matched terms are wrapped with __fstr_S__ / __fstr_E__ markers — strip before display.

Commands cheatsheet

Run four-star <cmd> --help for full options.

four-star self                                   # auth preflight
four-star -f json list --newer D --older D --limit N
four-star -f json get --id <recording_id>        # single meeting
four-star -f json search --q "<text>" --limit N  # full-text across chat + transcript

Output format flag (-f json or -f tab) must come before the subcommand. --limit goes after.

JSON shapes

Meeting (from list / get):

name, recording_id, recording_external_id, recording_owner_id,
created_at, external_modified, downloadable, shared, links
links: {download, drive, embed, thumbnail}

Optional, only present when populated:

chat_id, chat_external_id, chat_owner_id,
transcript_id, transcript_external_id, transcript_owner_id,
clean_name, indexed, recording_target_id

Search result (search):

name, date, attendees[], links
transcript_matches?: { blocks: [{ offset, lines: [{ speaker, line }] }] }
chat_matches?: ...

Known issues (do not fix)

list --name <str> returns nothing. Name filter is broken; find.sh pulls JSON by date range and filters client-side.
search sorts ascending by date with no reverse option and no date filter. Top hits for common titles are always from the earliest indexed year. Use list (or find.sh) with date bounds for "latest X" queries; use search only when you need transcript/chat content matches.
No four-star endpoint returns full transcript text. Only per-query snippets via search's transcript_matches. Meeting.links exposes the recording video, not the transcript document. Work around this by pulling the transcript Doc directly via the Google Drive MCP (see step 3).
Drive download links in Meeting.links require a Google session, not the four-star token. They 302 to accounts.google.com. Use the Drive MCP instead of fetching these URLs with curl.
JSON output omits null optional fields. Absence of transcript_external_id in the output means "no transcript indexed for this meeting," which is common — only a small fraction of recent meetings have indexed transcripts.

Source

Source lives at ~/oxide/four-star. OpenAPI spec at ~/oxide/four-star/meeting-api-spec.json is the reference for API shapes. Available endpoints: /meeting, /meeting/{id}, /search. Nothing else exposes transcripts.