ragmcp has 8 tools with mostly complete schemas and descriptions in French. Naming follows verb_noun patterns (search_documents, ingest_document, delete_source). However, descriptions lack LLM-optimization guidance (e.g., 'Recherche dans les documents indexés' is functional but generic). Parameter descriptions are present but sparse. The visual_ingest tool has NO visible input schema in the provided code. Error handling and recovery guidance are absent. Output schemas are not documented. Composition is reasonable (separate ingest/search/delete tools), but parameter relationships (e.g., session_id across tools) lack explicit documentation.

• Expand tool descriptions to 80-150 chars. Include WHAT the tool does, WHEN to use it (vs similar tools), and WHAT it returns. E.g., 'Search indexed documents by semantic or lexical similarity. Use for retrieval-augmented generation. Returns ranked chunks with source metadata and relevance scores.'
• Document output schemas for all tools. Specify field names, types, and whether results are paginated. E.g., search_documents returns {chunks: [{text, source_id, score, metadata}], total_count, next_cursor}.
• Add error handling guidance to destructive tools. E.g., delete_source: 'Permanently removes source and all chunks. This action cannot be undone. Verify source_id before calling.'
• Clarify session_id semantics. Document whether it enables multi-turn memory, whether it is required, and how it interacts with retrieval_mode and filters.
• Add input validation descriptions. E.g., top_k: 'Number of chunks to return (1-100, default 5). Higher values increase latency and token cost.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry