sync-transcripts

Sync Transcripts from Google Drive

Pull any meeting transcripts from Google Drive that aren't already in transcripts/, extract the transcript-only portion, and save them as markdown matching the existing format.

Default lookback is 3 days. If the user gives a date range or different lookback, use that.

Step 1 — Search Google Drive

Use mcp__claude_ai_Google_Drive__search_files with excludeContentSnippets: true (snippets blow up the output).

Query template — adjust modifiedTime to today minus the lookback window:

(title contains 'transcript' or title contains 'Transcript' or title contains 'Meeting' or title contains 'Notes by Gemini' or title contains 'Recording') and modifiedTime > 'YYYY-MM-DDT00:00:00Z'

Step 2 — Cross-reference against transcripts/

List the local folder. For each Drive file, check whether a local .md already covers that meeting (match on date + topic, not exact title — local names are abbreviated, e.g. Drive's BD AI System - India Country Partner Session - 2026/04/22 14:23 - Notes by Gemini → local 2026-04-22_BD_AI_India_Country_Partner.md).

Report the missing files to the user before downloading. If two Drive files map ambiguously to two local files (same date, same topic), flag it rather than guessing.

Step 3 — Confirm before downloading

Ask the user which missing transcripts to pull. Do not auto-download all of them.

Step 4 — Fetch and extract

For each confirmed file, call mcp__claude_ai_Google_Drive__read_file_content with the fileId.

The Drive file contains both Notes and Transcript sections. Extract only the transcript: everything from the ## <title> - Transcript heading down to (and including) ### Transcription ended after HH:MM:SS. Drop the Gemini disclaimer line that follows.

After the tool call, choose the right path based on what came back:

Path A — Content returned inline

The full file content is in the tool result directly. Apply the cleaning rules below, then use the Write tool to save the file. Do not write a script.

Path B — Content saved to a file (too large for inline)

The tool result says the output was saved to a path like:

<YOUR_HOME>/.claude/projects/<PROJECT_HASH>/tool-results/mcp-claude_ai_Google_Drive-read_file_content-*.txt

Run the extraction script:

python3 ".claude/scripts/extract_transcript.py" "<saved_tool_result_path>" "<output_transcript_path>" "<MMM D, YYYY>"

The script handles JSON parsing, transcript extraction, and cleaning automatically. See .claude/scripts/extract_transcript.py for full details.

Step 5 — Clean the content

(Only needed for Path A — Path B handles this automatically via the script.)

The raw content has JSON escapes and markdown bold that don't match the existing format. Apply these substitutions:

• **Speaker:** → Speaker: (strip bold from speaker labels)
• \& → &
• \* → *
• \[ → [
• \] → ]
• Preserve trailing two-space line breaks ( at line ends) — they render as line breaks in markdown

Step 6 — Save the file

Filename: YYYY-MM-DD_Snake_Case_Topic.md in transcripts/. Strip "BD AI System -", "Notes by Gemini", and timestamps from the title; keep it short and topic-focused. Examples:

• Drive: BD AI System - Director of Technical Services & Projects Session - 2026/04/24 11:00 - Notes by Gemini
• Local: 2026-04-24_BD_AI_Director_Technical_Services.md

File structure (match existing files exactly):

MMM D, YYYY
## <Original meeting title> - Transcript

### 00:00:00

Speaker: line one  
Speaker: line two  

### 00:01:29

Speaker: ...

Note: file starts with a blank line, then the date, then the ## title.

Step 7 — Update INDEX.md

For each newly saved transcript, add a row to the matching table in INDEX.md. Pick the section by meeting type:

Section	When to use
Stakeholder Interviews — Kerten Team	1:1 sessions with a Kerten employee (e.g. BD AI System - <Role>)
Intelligence Hub Sessions	Lead sourcing / market intelligence sessions
Kickoff & Progress Meetings	Stream kickoffs, weekly progress reviews
Technical Design Sessions	Architecture / workflow / KH Technical sessions
Vela Internal Check-ins	`Check-in

Use — in the Summary column (the /summarize-transcripts command picks up these pending rows). Match the column structure of the target table — they differ between sections. Date format: MMM D (e.g. Apr 24). Insert the row in date order.

Step 8 — Report back

Summarize what was added: filename(s), date(s), approximate duration (from the final ### Transcription ended after timestamp), and which INDEX.md section(s) you updated.