crawl-mcp has 8 tools with complete input schemas and descriptions. Naming follows verb_noun convention (crawl_*, search_*). Descriptions are detailed (100-400 chars) and explain WHEN to use each tool vs alternatives, which is excellent for LLM selection. However, output schemas are documented only in docstrings, not in formal JSON Schema. Parameters lack enums for constrained values (e.g., region, safesearch, timelimit formats are described but not formally constrained). Error handling is minimal, no guidance on recovery or retry logic. Tool composition is good: crawl_single/batch/site are distinct; search_* tools are separate from crawl tools. Security: no secrets exposed as parameters (llm_config is optional and server-side validated). Overall: solid definitions with room for formal schema rigor and error guidance.

• Add formal JSON Schema definitions for all output types. Document response structure with field types, required fields, and examples. Use tool.outputSchema or equivalent in FastMCP.
• Convert region, safesearch, timelimit, size, color, type_image, layout parameters to enum constraints in schema. Remove free-form string descriptions and replace with enum values.
• Add error handling guidance to all tool descriptions: 'On timeout (>60s), retry with reduced concurrent or smaller batch. On 'content too short', try enhanced=True or prefer_fast=False.'
• Correct search_images risk classification: mark as READ_ONLY for search operations; only mark WRITE if download=true or analyze=true.
• Document timeout behavior in tool descriptions: 'This tool has a 75s timeout. If exceeded, returns error with retry guidance. Reduce concurrent or batch size to avoid timeouts.'

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

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