Single tool 'find_mcp' has a well-structured schema with proper Zod validation, clear descriptions for all parameters, and explicit constraints (minLength, maxLength, min, max). Description is concrete and action-oriented (194 chars), explaining what it returns and when to use it. Parameter descriptions are detailed and actionable. However, output schema is not documented in the visible code, the Hit interface exists but is not exposed as part of the tool definition or handler return type annotation. Tool name is verb-noun compliant. No critical gaps, but output documentation would elevate this to 80+.

• Document the output schema explicitly. Add a return type annotation to the handler or include a schema field in the ToolDef that describes the Hit structure (name, description, source, remote_url, use_count, updated_at, etc.).
• Add an example to the tool definition showing a realistic invocation: { query: 'postgres', limit: 5, remote_only: true, task_context: 'I need to query a PostgreSQL database from my agent' }.
• Enhance the description with error recovery guidance: 'If no results are found, try a shorter query (1-2 keywords) or check if the upstream registries are available.'
• Verify that task_context is properly required in the final schema after withTaskContext() is applied, and document this in the tool description.
• Consider adding a 'sort_by' parameter (enum: popularity, recency, relevance) to give agents more control over result ranking, since the description mentions ranking but the parameter is not exposed.

Inferred effective spec: 2025-06-18+.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained