anyrepair exposes 12 tools with minimal descriptions and basic schemas. Most tool names follow verb_noun convention (repair_json, repair_yaml, etc.), which is good. However, descriptions are extremely terse (10-40 chars), providing almost no context for LLM selection. All tools accept a single 'content' parameter with minimal detail. The 'repair' tool lacks documentation of its auto-detection behavior. The 'validate' tool is the only one with a more complex schema (enum for format), but still lacks depth. No output schemas are documented, LLMs cannot predict response structure. No error handling guidance provided. Tool definitions appear to be derived from src/mcp_server.rs and the example file, but full schema construction logic is not visible in the provided source, making it difficult to verify exact parameter constraints. The server is functional but well below production quality for agent integration.

• Expand each repair_* description to 80-150 chars. Example for repair_json: 'Repair malformed JSON: fix trailing commas, quote keys, convert single quotes to double. Returns repaired JSON string. If unparseable, returns error with position hint.'
• Expand 'repair' description to clarify auto-detection: 'Auto-detect content format (JSON, YAML, XML, CSV, etc.) and apply appropriate repair. Use this when format is unknown; use repair_json/repair_yaml/etc. when format is known for better performance.'
• Add formal output schema documentation. Example: 'Returns object with fields: repaired (string, repaired content), format (string, detected format), confidence (number 0-1), errors (array of error objects with line, column, message).'
• Document parameter constraint for 'content' parameter: 'Content string to repair. Max 10MB. Must be text (UTF-8). Binary content should be base64-encoded.'
• Add error handling section to each description: 'On repair failure, returns error object: {error: string, reason: enum(unparseable|ambiguous|truncated), suggestion: string}. Unparseable indicates format detection failed, try explicit repair_* tool.'
• Document size/complexity limits: 'Repair is optimized for documents <5MB. Larger inputs may timeout; break into chunks and repair separately.'
• Add dependency hint to 'repair': 'If you know the format, prefer repair_json, repair_yaml, etc. for better performance and more detailed error messages.'
• For 'validate' tool, expand description: 'Validate content against format syntax rules. Returns {valid: boolean, errors: [array of syntax errors with line/column], warnings: [array of non-fatal issues]}. Does not repair, use repair_* tools to fix issues.'