Single tool with visible registration in src/server.js. Tool has a description and input schema with Zod validation, but critical quality gaps prevent a higher score. Description is adequate (89 chars) but lacks actionable guidance on when to use which strategy. Input parameters have type constraints (enum for strategy, integer bounds for maxBatches) and descriptions, but descriptions are minimal (10-30 chars) and lack formatting/validation hints. Output schema is documented but contains z.any() for 'matches' field, indicating incomplete typing. No error recovery guidance, no dependency hints, no examples of output structure. The tool implements multiple search strategies but does not explain the trade-offs or when to prefer one over another, an LLM cannot infer this from the current description. Tool name 'trocco_batch_search' follows verb_noun convention and is clear.

• Replace z.any() in output schema's 'matches' field with explicit object schema: z.array(z.object({ id: z.string(), name: z.string(), description: z.string().optional(), ..., config_details: z.record(z.any()).optional() }))
• Expand strategy enum descriptions: exhaustive_scan: 'Scan all definitions systematically via pagination (slower, comprehensive)'; keyword_chunks: 'Search using substrings of your term (faster for partial matches)'; etc.
• Add parameter description examples and constraints: 'maxBatches controls number of API pages to fetch; typical page size is 100 items, so 10 batches = up to 1000 scanned'
• Document the output 'matches' structure in tool description: 'Returns matching job definitions with id, name, description, and extracted S3/Snowflake/BigQuery configuration details'
• Add error recovery guidance to tool description: 'If search fails or times out, try a smaller maxBatches or simpler searchTerm. If no matches found, try recent_first strategy.'
• Document TROCCO_API_KEY injection in tool description: 'Requires TROCCO_API_KEY environment variable to be set on server startup. Authentication is automatic.'
• Add note on searchProgress field format: 'searchProgress is a human-readable string summarizing batches searched, items scanned, and matches found'
• Implement structured error responses that indicate failure mode (timeout, auth_error, rate_limited, no_matches) rather than silent failures