A Model Context Protocol (MCP) server that provides access to Crunchbase data for AI assistants
Scoring was not performed
Output schemas are not documented. While input schemas are present, the tool descriptions do not specify what fields the API returns, what format data takes, or which fields are useful for downstream chaining. LLMs cannot plan multi-step workflows without knowing response structure.
Parameter constraints are missing. 'limit' has no min/max specified (baseline: 1 - 100). 'status' accepts free-form strings instead of an enum (valid values: 'active', 'closed', should be enum). 'founded_after' and 'founded_before' accept strings but do not specify format (YYYY-MM-DD is mentioned in description but not in schema).
Error messages lack recovery guidance. When resource URIs fail to match (lines 120 - 134), the server returns 'Invalid URI' but does not suggest valid alternatives or how to proceed. LLMs receive dead-end errors instead of actionable next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | F | 32 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 57 | - | v1 |
Tool descriptions are minimal (65 - 70 chars). Descriptions answer WHAT but not WHEN or WHY. E.g., 'Get detailed information about a specific company' does not explain when to use this vs search_companies or what additional context it provides. Baseline for good descriptions is 50 - 200 chars with intent/usage guidance.
Parameter descriptions lack format/constraint details. 'company_name_or_id' accepts either name or UUID but does not specify the format or priority (is 'Apple' or 'uuid-12345' preferred?). This invites LLM confusion and potential lookup mismatches.