Gong MCP server implementation for accessing Gong call data and transcripts via the Gong API v2
The Gong MCP server defines 2 tools with explicit schemas and descriptions, but several quality gaps prevent a higher score. Both tools have clear action verbs and moderately helpful descriptions (130 - 175 chars), meeting the basic floor. However, parameters lack per-parameter descriptions, critical constraint annotations are missing, output schemas are undocumented, and error handling provides no recovery guidance. The implementation shows competent API integration but falls short of production-grade tool quality, typical of a C-/D+ tier community server.
List Gong calls with optional date range filtering. Returns call details including ID, title, start/end times, participants, and duration.
Retrieve transcripts for specified call IDs. Returns detailed transcripts including speaker IDs, topics, and timestamped sentences.
Missing per-parameter descriptions. 'fromDateTime' and 'toDateTime' in list_calls lack explanations of their purpose, format validation rules, or dependencies.
No output schema documentation. Neither tool documents what fields are returned or their types. LLMs cannot predict response structure or plan downstream operations.
Missing format constraints on datetime parameters. 'ISO format (e.g. 2024-03-01T00:00:00Z)' is in the description but not enforced via JSON Schema pattern or minLength/maxLength. LLMs may pass malformed dates.
No pagination support. list_calls returns all matching calls with no limit, offset, or cursor. Large date ranges could return hundreds of calls, exhausting context windows.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 57 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 46 | - | v1 |
Error handling provides no recovery guidance. CallToolRequestSchema handler catches errors but returns generic 'Invalid arguments' or unstructured stack traces instead of actionable messages like 'Invalid callIds: must be non-empty array of strings'.
callIds parameter in retrieve_transcripts has no constraint on array size. An LLM could pass 1000+ call IDs in a single request, causing API timeout or rate-limit penalties.
Tool descriptions lack 'WHEN to use' guidance. 'Retrieve transcripts for specified call IDs' does not explain when an agent should call this vs. list_calls, or what a user would ask for that justifies the call.