This MCP server has three tools with minimal definitions and no visible input schemas. All tools have extremely brief descriptions (under 20 chars for two of them), no parameter descriptions, and no documented output schemas. The tools appear to be wrappers around a 'skill-watch' configuration system, but the MCP definitions lack the structure and clarity required for reliable LLM integration. Tool names are moderately clear (skill_watch_discover, skill_watch_sources, skill_watch_check) but descriptions fail the 20-character minimum threshold, triggering hard scoring rules.

• Add detailed output schemas to all three tools. Document the structure of returned objects (e.g., what fields does skill_watch_discover return? Is it a list of references with name/url pairs?). Include example responses.
• Expand skill_watch_check description to clarify: 'Returns a diff report comparing the current state of the specified source (identified by source_id) against its previous state. Useful for monitoring changes in skill-watch targets.' Include what format the diff takes (JSON, unified diff, etc.).
• Add explicit descriptions to the source_id parameter in skill_watch_check: 'The unique identifier of the source to monitor (e.g., "github-docs", "internal-wiki"). Use skill_watch_sources to list available source IDs.'
• Document the expected result structure for skill_watch_discover and skill_watch_sources in their descriptions. E.g., 'Returns an array of reference objects, each containing {id, name, url, documentation_link}.' This guides LLM expectations.
• Add error handling guidance to tool descriptions. E.g., 'If source_id is not found, returns a 404 error with available source IDs. If configuration is invalid, returns a validation error with details.' This helps LLMs recover from failures.
• Clarify the relationship between tools. E.g., add to skill_watch_discover: 'Call skill_watch_sources first to see available sources, then skill_watch_check to monitor a specific source.'
• Specify limits and constraints in descriptions. E.g., 'Returns up to 100 most recent changes for skill_watch_check. Use pagination parameters (if supported) for longer history.' Add pagination support if missing.