TheHood exposes 36 tools with consistent naming (verb_noun pattern: thehood_*) and basic schemas. However, descriptions are generic and parameter documentation is sparse. Most tools lack actionable error guidance, output schemas are not documented, and many parameters have no descriptions. The codebase shows tool definitions in src/mcp/tools.ts with ToolDefinition and ToolResult types, but descriptions are minimal (10-50 chars) and parameter details are often missing. No evidence of pagination, idempotency hints, or recovery guidance. This is a functional but immature tool suite typical of early-stage MCP servers.

• Document output schemas for all tools. For list_* tools, specify: total count, items array, pagination cursor/offset, and per-item field structure. Example: 'Returns {runs: [{id, goal, status, createdAt}], total: number, nextCursor?: string}'.
• Add parameter constraints to descriptions. For 'maxBytes', 'maxDepth', 'maxCycles', 'maxItems': specify min/max ranges and defaults. E.g., 'maxBytes: Maximum bytes to read (1 - 1048576, default 65536)'.
• Add error recovery hints to tool descriptions. For write operations (create_run, abort_run, summon_agent), document: 'Returns error if run not found or already in terminal state. Call list_runs() to verify run exists before retrying.'
• Implement pagination for list_* and search_* tools. Add 'limit' (default 20, max 100) and 'offset' or 'cursor' parameters. Return total count and next_cursor in response so agents can iterate without guessing.
• Document tool chaining requirements. For each tool, list required input fields and guaranteed output fields. E.g., 'get_run returns {id, goal, status, artifacts: [{ref, kind}]}, use artifact refs with read_artifact().' This enables agents to compose tools correctly.