FlintTrade exposes a single tool 'mcp:stdio' with minimal definition. The tool name violates naming conventions (should start with action verb like 'get_' or 'list_'), lacks parameter schema entirely, and has only a generic 26-character description. No input/output schemas are visible in the provided source. The tool is inferred from package.json metadata rather than explicit registration code. This is a documentation/metadata server, not a functional tool suite.

• Rename tool from 'mcp:stdio' to an action verb: 'get_site_documentation', 'search_site_content', or 'list_documentation_pages'. This enables LLMs to infer intent from the name alone.
• Add a 50 - 200 character description explaining: (1) what content is exposed, (2) when to call this tool (e.g., 'when user asks about FlintTrade features or setup'), (3) what structure is returned (e.g., 'returns markdown pages with title, url, and body').
• Define explicit input parameters with types and descriptions. E.g., 'query' (string, required, 'search term or topic'), 'limit' (integer, optional, default 10, 'max results to return'), 'format' (enum: markdown|html|plaintext, 'output format').
• Document the output schema with typed fields. E.g., 'returns array of {title: string, url: string, body: string, relevance_score: number}'. Include pagination (total_count, next_cursor) if results can exceed limit.
• Add error handling guidance. E.g., 'If no results found, return empty array with message: "No documentation matches query. Try broader terms or browse site directly."' This prevents LLM confusion on empty results.