Three tools with clear verb-noun naming (local_research, local_status, local_recover). All have descriptions (10 - 200 chars, within baseline). Input schemas present with type definitions and descriptions. However, output schemas are not documented, LLMs cannot predict response structure. Parameter descriptions lack detail on constraints, formats, and error recovery paths. No tool annotations (readOnlyHint/destructiveHint). Error handling guidance is minimal. Composition is reasonable: tools are single-purpose and chainable via context passing.

• Document output schemas for all three tools. Example for local_research: 'Returns {"evidence": [{"title": str, "url": str, "snippet": str}], "warnings": [str], "duplicate_tool_call": bool}'. This lets LLMs extract fields reliably.
• Add enum constraints to parameters. Change 'freshness' description to: 'Freshness constraint: one of day, week, month, year, or null (no constraint). Default: null.' Change 'mode' to: 'Research mode: one of auto, fast, comprehensive. Default: auto.'
• Add tool annotations via FastMCP. Mark local_research and local_status with readOnlyHint=true. Mark local_recover with destructiveHint=true. This signals to agents which calls are safe to retry.
• Add recovery guidance to descriptions. local_research: 'If search backend is unavailable, call local_recover() first.' local_status: 'Returns diagnostics; if status is degraded, call local_recover() to restart services.'
• Expand parameter descriptions. 'include_logs' → 'Include recent MCP startup logs in diagnostics (default: true). Set to false to reduce output size.' 'include_api' → 'Start the FastAPI/UI service after Docker/SearXNG (default: false). Set to true to enable web UI access.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry