The OpenRewrite MCP server defines 3 tools with explicit schemas and descriptions visible in mcp-server/src/server.py. Tool naming follows verb_noun convention (test_connection, find_recipes, get_recipe) with action-oriented verbs. Descriptions are present and reasonably detailed (194 - 250 chars), within the production baseline of 34 - 392 chars. However, there are significant gaps in schema completeness, parameter descriptions, and output documentation. The find_recipes tool has a sophisticated oneOf schema for flexible input (string or array), but other tools lack proper output schema documentation. Error handling and recovery guidance are absent. Parameter validation constraints are present but descriptions could be more explicit about formats and constraints. The test_connection tool is trivial and contributes minimal value.

• Document output schemas for all tools. For find_recipes, explicitly define: [{ recipe_id: string, name: string, description: string, relevance_score: number }, ...]. For get_recipe, define: { recipe_id: string, name: string, description: string, documentation: string, examples: [...], config_options: {...}, ...}.
• Expand get_recipe description to 100 - 150 chars and include: 'Use after find_recipes to fetch full documentation, examples, and configuration for a specific recipe. Requires exact recipe_id from find_recipes results.'
• Add explicit format and semantic documentation to min_score: 'Minimum relevance score (0.0 - 1.0, where 1.0 is perfect semantic match). Default 0.5 filters out low-confidence results; increase to 0.7+ for strict matching.'
• Remove test_connection or repurpose it as a health check that validates database connectivity and semantic embedding model availability, returning structured status with timestamps.
• Document failure modes with recovery guidance: 'If find_recipes returns zero results, try broader keywords or simplify the intent. If get_recipe fails with 'Recipe not found', verify the recipe_id from find_recipes output.' This implements pattern:recovery-guide.
• Add output schema documentation to find_recipes describing single vs. multi-query behavior: 'Multi-query mode uses Reciprocal Rank Fusion to combine results; recipes ranked higher by any query bubble to the top, improving recall for ambiguous intents.'
• For find_recipes, clarify pagination: 'Returns up to `limit` results. If total_available > limit, results are ranked by relevance score (highest first). No cursor-based pagination; for exhaustive search, increase limit parameter.'

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

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