Strong foundation with 14 well-named tools following verb_noun patterns (search_law, get_article, list_laws, verify_citation, find_references, search_program, get_program, list_open_programs, match_programs, data_status, delegation_map, startup_stage_guide, check_effective_date, sync_programs). All tools have descriptions (avg ~180 chars, within 10-1024 baseline). Input schemas are visible and properly typed with descriptions for most parameters. However, output schemas are NOT documented in the source code, only input schemas are shown. Error handling is minimal (basic validation functions exist but no recovery guidance in tool descriptions). Tool composition is excellent: each tool has a single responsibility, and the server instructions clearly guide when to use each tool. Parameter naming is consistent (source, article, query, limit, status). Missing: output schema documentation, error recovery guidance, and per-tool error classification.

• Document output schemas for all 14 tools. For each tool, add a 'Returns:' section in the docstring listing field names, types, and descriptions. Example: 'Returns: [{source: str, article: str, snippet: str, score: float}]'.
• Enhance error handling descriptions. For tools with validation (search_law, match_programs, etc.), add guidance: 'If limit is invalid, the tool returns {status: "invalid_input", field: "limit", message: "..."}. Correct the value and retry.'
• Expand data_status and delegation_map descriptions to 100+ chars. Explain what fields are returned and when to call. Example: 'Returns sync timestamps, data freshness, and last update source. Call this to verify data is current before relying on search results.'
• Add explicit state-modification warning to sync_programs: 'This tool fetches and updates K-Startup program data from the remote API. It modifies server state and should only be called when data refresh is needed. Not idempotent, repeated calls may trigger duplicate API requests.'
• Add tool annotations to FastMCP registration. Use @mcp.tool(readOnlyHint=True) for search/get/list/verify/find/check tools, and @mcp.tool(destructiveHint=True) for sync_programs to signal safety properties to agents.