Local MCP server that exposes read-only tools for browsing and searching meeting transcripts stored by Meeting Transcriber macOS app. Allows external MCP clients (Claude Code, etc.) to list, retrieve, and search transcripts without touching the app's internal state.
Four well-named, read-only tools with complete input schemas and clear descriptions. All tools follow verb_noun naming (list_, get_, search_). Descriptions are substantive (100 - 250 chars), explaining WHAT each tool does and WHEN to use it. Input schemas are properly typed with required fields marked. Tool annotations (readOnlyHint, idempotentHint) are present and correct. Key gaps: output schemas are not documented in the tool definitions, LLMs cannot see what fields to expect from responses. Parameter descriptions are minimal (e.g., 'Transcript id from list_transcripts' lacks format/constraint details). Error handling is mentioned in get_summary description but not systematized across all tools.
Return the LLM-generated summary, generation model and timestamp for the given transcript id. Returns an error result if the transcript hasn't been summarized yet.
Return the full markdown transcript for the given id (as listed by list_transcripts).
List transcripts stored by Meeting Transcriber. Returns a JSON array sorted newest-first, each entry containing id, title, ISO-8601 date, durationSeconds, language, and hasSummary. Pass an optional `query` to substring-match the title.
Full-text search across every stored transcript. Matches the query (case-insensitive) against transcript titles and every spoken segment, returning a JSON array (newest-first) of matching transcripts — each with id, title, ISO-8601 date, whether the title matched, and up to `maxHitsPerTranscript` matched segments (time in seconds, speaker, text). Use this to answer questions like "what did we decide about X?" across all meetings, then `get_transcript` to read one in full.
Output schemas not documented. Tool definitions lack descriptions of response structure (fields, types, format). LLMs cannot plan downstream calls or extract data without knowing what fields to expect.
Parameter descriptions lack format/constraint details. 'Transcript id from list_transcripts' does not specify format (UUID? alphanumeric?), length, or validation rules. 'limit' and 'maxHitsPerTranscript' lack min/max bounds.
Error handling not systematized. Only get_summary mentions error behavior ('Returns an error result if not summarized'). Other tools lack guidance on what errors can occur and how to recover (e.g., 'id not found' → try list_transcripts).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 67 | 2026-07-28+ | v2 |
search_transcripts lacks pagination guidance. Description mentions 'default 20' and 'default 5' but does not explain how to fetch next page or total count. Large result sets risk context window exhaustion.
Input schema structure uses non-standard nesting. Properties are wrapped in .object([...]) with 'type' as a key, rather than standard JSON Schema format. This may confuse schema validators or clients expecting canonical JSON Schema.