MCP server validation CLI — fast pre-publish diagnostic for Model Context Protocol servers. Schema validation, tool/resource/prompt health checks, scorecard reporting, CI-ready exit codes.
Two tools with clear, verb-prefixed names (probe_server, score_server) and solid descriptions (150+ chars each). Input schemas are well-defined with Zod, all parameters typed and described. However, output schemas are not formally documented, responses are returned as text + structuredContent but the shape of structuredContent is not declared. Error handling returns actionable messages but lacks recovery guidance. Tool composition is clean (each does one thing), but the server itself is a meta-tool for inspecting other servers, so traditional composition patterns don't apply.
Inspect an MCP server: list its tools/resources/prompts, call them, and report schema and protocol-compliance issues.
Score an MCP server 0–100 on publishability (description quality, enum/schema shape, mutation legibility, distribution metadata) with a letter grade — run this before publishing a server.
Output schema not formally documented. Tools return structuredContent as Record<string, unknown> with no declared shape. LLMs cannot plan downstream operations or validate response structure.
Error handling returns actionable messages but lacks recovery guidance. Errors do not suggest next steps (e.g., 'Try calling probe_server first to inspect the target').
No parameter validation or constraint documentation beyond type. E.g., timeout is a number with no min/max bounds; headers array has no format guidance.
Tool descriptions do not explicitly state what data is returned or how to interpret results. E.g., probe_server returns a summary object but the description does not list expected fields.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2025-06-18+ | v2 |