Unofficial MCP server for Mistral AI docs and API reference. Provides search and filesystem access to Mistral documentation.
Two well-named, read-only tools with clear descriptions and proper Zod schemas. search_mistral_docs has good parameter constraints (limit 1-20). Both tools include helpful usage guidance in descriptions. However, output schemas are not formally documented, responses are JSON strings without declared field types. Parameter descriptions are present but could be more explicit about return structure and error cases. No error handling guidance for agents (e.g., what to do if search returns empty, or if filesystem command fails).
Read-only shell over a virtualized filesystem of Mistral's docs. Supports rg (regex), find (glob), cat (read page), ls (list directory). Good for exact-string lookups the search index misses, reading a full page after search_mistral_docs, or exploring the doc tree. Output capped at 30 KB.
Search Mistral AI's docs and API reference. Use for any Mistral question: models, the API, fine-tuning, agents, embeddings, OCR, pricing, SDKs. BM25 with fuzzy matching. Returns ranked pages with title, source URL, doc_path (feed into query_mistral_docs_filesystem), and a snippet. Start here; fall back to query_mistral_docs_filesystem for exact strings or full-page reads.
Output schemas not documented. Tools return JSON strings without declared field types (title, source_url, doc_path, snippet, score for search; command output for filesystem). LLMs cannot plan downstream operations or validate response structure.
No error handling guidance. Descriptions do not explain what happens on failure (empty results, invalid regex, file not found, output truncation at 30 KB). Agents cannot recover from errors or know when to retry.
query_mistral_docs_filesystem command parameter lacks format constraints. Description lists supported commands (rg, find, cat, ls) but does not validate input or explain error cases (invalid regex, glob syntax errors, path traversal attempts).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
search_mistral_docs limit parameter defaults to 8 but description says 'Default 8, max 20', default is not enforced in schema (marked optional). If omitted, behavior depends on implementation, not schema.
No pagination or result limits documented for filesystem commands. Output is capped at 30 KB but agents are not told how to handle truncation or request next page.