Three tools with minimal schema definition and generic descriptions. get_version and list_available_mcps have empty input schemas ({}), which is correct for parameterless tools, but load_mcp_tools has only a single string parameter with basic description. Descriptions are present but lack actionable detail on when/why to use each tool or what they return. No output schemas documented. No error handling guidance. Tool names follow verb_noun pattern (acceptable) but descriptions do not explain return types or downstream usage.

• Add explicit output schema documentation for each tool. E.g., get_version returns {version: string}, list_available_mcps returns {servers: [{name: string, description: string, status: 'available'|'missing_env'}]}, load_mcp_tools returns {tools: [{name: string, description: string, inputSchema: object}]}.
• Expand load_mcp_tools description to include: 'Returns a list of tools available from the specified MCP server. Each tool includes its name, description, and input schema. If the server is not found or fails to initialize, returns an error with available server names.'
• Add enum constraint to load_mcp_tools mcp_name parameter: derive valid values from MCP_SERVERS keys and document in the parameter description.
• Document error cases and recovery paths: 'If mcp_name is invalid, call list_available_mcps() to see available servers. If the server fails to initialize, check that required environment variables are set (see list_available_mcps output for warnings).'
• Return structured JSON from list_available_mcps instead of a formatted string. This allows LLMs to parse and reason about server metadata programmatically, and enables filtering/sorting.