Two tools with minimal documentation and no structured output schemas. Both tools have descriptions (French-language, 80-120 chars), but lack parameter descriptions, output schema documentation, and error handling guidance. No input validation, no pagination support, no error recovery patterns. Tool names are reasonably clear (search_*, ingest_*) but descriptions are generic. Hardcoded API credentials and bucket IDs in code violate secret injection pattern. No tool annotations (readOnlyHint/destructiveHint). Schemas are present but minimal (single string params, no constraints). This server falls well below production baseline.

• Add parameter descriptions to both tools. For search_doc_for_rag_context: 'query (string): Natural language search query. Max 500 chars. E.g. "machine learning best practices"'. For ingest_documents: 'local_file_path (string): Absolute or relative path to a PDF file. Must exist and be readable. E.g. "/data/docs/report.pdf"'.
• Document output schemas. search_doc_for_rag_context should return {"content": string, "source_count": int, "confidence": float} instead of raw text. ingest_documents should return {"status": "success"|"error", "file_name": string, "estimated_availability_minutes": int, "error_message": string|null}.
• Move hardcoded bucket IDs (17221, 17279) to environment variables or configuration file. Accept optional bucket_id parameters in tool signatures to allow multi-bucket support.
• Add try-catch blocks with actionable error messages. E.g. 'File not found: /path/to/file. Check the path and try again.' or 'Search failed: GroundX API returned 429 (rate limited). Retry in 60 seconds.'
• Add tool annotations via FastMCP. Mark ingest_documents with destructiveHint=true and search_doc_for_rag_context with readOnlyHint=true so agents understand operation safety.