CogniResearch has 12 tools with basic schemas and descriptions, but significant gaps prevent production use. Most tools have descriptions (10/12), but they are generic and lack actionable context. Input schemas are present but incomplete: many parameters lack type constraints, enums, or validation guidance. No output schemas documented. Error handling is minimal, tools return raw error strings without recovery guidance. Parameter descriptions are sparse (e.g., 'The search query' for web_search lacks format/length constraints). No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels. The clear_memory tool is destructive but has no confirmation pattern. Overall, tools follow basic naming conventions (verb_noun) but lack the depth required for reliable LLM reasoning.

• Add output schemas to all tools. Document return types as JSON Schema with field descriptions. E.g., web_search should return {results: [{title: string, url: string, snippet: string}], total_count: integer}.
• Implement confirmation pattern for clear_memory: add a 'confirm' boolean parameter (default false) and require explicit confirmation before deletion. Return 'Confirmation required. Call again with confirm=true to proceed.'
• Enhance parameter descriptions with constraints: query_database should specify 'SELECT-only queries; no INSERT/UPDATE/DELETE. Max 10,000 rows returned.' search_files should document glob pattern syntax (e.g., '*.txt', '**/*.py').
• Add tool annotations to all tools via MCP schema. Mark write_file and clear_memory as destructiveHint=true. Mark all read tools as readOnlyHint=true. Mark idempotent tools (e.g., read_file) as idempotentHint=true.
• Improve error messages with recovery guidance. E.g., 'TAVILY_API_KEY not set. Configure via .env file or environment variable.' For 'User not found' errors, suggest 'Try search_users() with a partial name.' or list available options.