Server has 6 tools with basic descriptions and some parameter schemas, but significant gaps in definition quality. All tools are READ_ONLY and return unstructured text (markdown strings) rather than structured JSON. Parameter descriptions exist but are minimal (10-50 chars). No output schemas documented. No error handling guidance. Tool names follow verb_noun convention (get_*, list_*, query, add_*) which is good, but descriptions lack actionable context for LLM selection. The query tool accepts optional 'sources' array but filtering logic is not visible in implementation. Overall pattern adherence is weak, most tools would score 35-50 individually.

• Convert all tool responses from markdown strings to structured JSON objects. E.g., get_projects should return {projects: [{id, name, status, description}], total: N} instead of formatted markdown.
• Document output schemas for all tools. Specify field names, types, and whether arrays are paginated. Include example responses.
• Expand tool descriptions to 100-200 chars. Add WHEN to use each tool and what it returns. E.g., 'Get all active projects with metadata. Use this to discover available projects; use query() for full-text search across project content.'
• Implement the sources filter in query() tool. Pass sources parameter to embeddings.search() to enable filtering by source name.
• Add error handling with recovery guidance. Return structured error objects: {error: 'not_found', message: '...', suggestion: 'Try calling list_sources() first'}. Distinguish retryable vs fatal errors.