mcp-search-proxy defines 4 tools with basic schemas and descriptions. All tools have input schemas with type definitions and descriptions present. However, descriptions are minimal (10-60 chars), falling below the 50-200 char LLM-optimized baseline. No output schemas are documented. Error handling is absent, tools do not guide recovery or categorize failures. Tool names use mcp_ prefix (non-standard verb_noun pattern). No parameter validation guidance, no enum constraints, no idempotent/destructive hints. Composition is reasonable (search → describe → call → refresh), but lacks chaining IDs and pagination support.

• Expand tool descriptions to 50 - 200 chars. Example: 'mcp_search: Search the cached upstream tool catalog using BM25 ranking. Returns tool names, descriptions, and relevance scores. Call this first to discover available tools before invoking mcp_describe or mcp_call.'
• Document output schemas for all tools. Example: mcp_search returns [{name: string, description: string, score: float, upstream_url: string}]. mcp_call returns {result: object, error?: string}.
• Rename tools to follow verb_noun pattern: search_tools, describe_tools, invoke_upstream_tool, refresh_catalog. Avoid mcp_ prefix or use it consistently with clear verbs.
• Add error handling and recovery guidance. Example: 'If mcp_call fails with "tool not found", call mcp_search(query) to discover the correct tool name. If upstream is unreachable, call mcp_refresh() to rebuild the catalog.'
• Add parameter validation rules to descriptions. Example: 'limit: integer, 1 - 25 (default 5). If omitted, defaults to 5. If > 25, capped at 25.' Add enum constraints where applicable (e.g., if mcp_call has a 'mode' param, list valid modes).

Inferred effective spec: 2026-07-28+.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained