mcp-uni exposes 3 core tools with basic schemas and descriptions. Tool names follow verb_noun convention (connect_mcp, disconnect_mcp, list_mcp), which is good. However, descriptions are minimal (10-30 chars), parameter descriptions are sparse or missing, and output schemas are not documented. The connect_mcp tool has a complex oneOf schema for transportConfig, but lacks guidance on when to use SSE vs stdio. Error handling is minimal, no recovery guidance or actionable error messages.

• Expand tool descriptions to 50-200 characters. Example: 'connect_mcp: Dynamically connect a remote MCP server (SSE or stdio) to this gateway. Use this to add new tools, prompts, and resources from external servers. Requires valid transport config (URL for SSE, command for stdio).'
• Add descriptions to all transportConfig sub-fields. For SSE: 'url: The server endpoint (e.g., https://example.com/stream). query: Optional query parameters. headers: Optional HTTP headers (e.g., Authorization).' For stdio: 'command: Executable path. args: Command arguments. env: Environment variables. cwd: Working directory.'
• Document output schemas explicitly. Example for list_mcp: 'Returns {content: [{type: "text", text: "List of MCP servers: server1, server2, ..."}]}'. For connect_mcp: 'Returns {content: [{type: "text", text: "Connected MCP server: <name>"}]} on success.'
• Add error handling with recovery guidance. Wrap mcpHost.addClient() in try-catch and return structured errors: 'Failed to connect to <name>: <reason>. Verify transport config (URL reachable, command executable, env vars set).' This guides LLM retry logic.
• Validate transportConfig at tool invocation time. Check that SSE URLs are valid URIs, stdio commands exist, and required fields are present. Return clear validation errors: 'Invalid transport config: missing required field "url" for SSE transport.' instead of generic failures.