MCP server for Fathom.video - Access AI meeting transcripts, summaries, and action items from Cursor IDE
Fathom-mcp has well-structured tool definitions with consistent naming (verb_noun pattern) and complete input schemas using Zod. All 7 tools have descriptions and typed parameters. However, descriptions are generic and lack LLM-optimization guidance (e.g., 'when to use this instead of similar tools'). Parameter descriptions are minimal (avg ~40 chars vs baseline 72). Output schemas are not documented, LLMs cannot predict response structure. Error handling is basic (isError flag only, no recovery guidance). No tool annotations (readOnlyHint/destructiveHint) despite clear read/write distinction. Missing dependency hints and chaining IDs in responses.
Export all meetings matching optional date filters to markdown files
Export a single meeting to a markdown file
Get action items from a specific meeting or recent meetings
Retrieve a specific meeting with optional transcript, summary, action items, and CRM matches
Get the full transcript for a specific meeting recording
List meetings with optional filters for date range, external participants, and limits
Output schemas not documented. LLMs cannot predict response structure or plan downstream calls. E.g., list_meetings returns markdown text only, no structured fields for recording_id, title, date that downstream tools (export_meeting, get_meeting) need.
Tool annotations missing. export_meeting and export_all_meetings are destructive (write to disk) but lack destructiveHint. list_meetings, get_meeting, get_transcript, search_meetings, get_action_items are read-only but lack readOnlyHint. Agents cannot distinguish safe from risky operations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Search meetings by participant emails, company domains, and teams
Parameter descriptions are minimal and lack LLM-optimization guidance. E.g., 'Maximum number of meetings to return (default: 10, use 0 for all)' does not explain when to use limit vs when to omit it. No dependency hints (e.g., 'If you only have a name, call search_meetings first').
Error handling is minimal. Only isError flag returned; no recovery guidance. E.g., 'Meeting with recording ID X not found' does not suggest calling search_meetings or list_meetings to discover valid IDs.
Response format is markdown text only. Agents cannot extract structured data (recording_id, title, date, participants) for chaining to downstream tools. Markdown parsing is error-prone and wastes tokens.