MCP Inspector is a meta-tool for inspecting other MCP servers. Tool definitions are present with basic descriptions and schemas, but fall short of production quality. Most tools lack depth in parameter descriptions and output schema documentation. Naming is clear and verb-based (list*, read*, call*, get*, set*), but descriptions are generic and under-optimized for LLM selection. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Error handling guidance is absent, tools do not explain recovery paths. The server exposes internal MCP protocol details (listTools, callTool) rather than abstracting them into domain-specific operations.

• Document output schemas for all list* tools. Define McpResponse structure explicitly: { tools: Tool[], resources: Resource[], prompts: Prompt[] } with field types and descriptions.
• Expand tool descriptions to 80-150 characters. Example: 'List all tools available on the connected MCP server. Call this first to discover what operations are possible, then use callTool to invoke them.'
• Add destructiveHint and idempotentHint annotations to callTool and setLoggingLevel. Mark callTool as potentially destructive (depends on the tool being called); mark setLoggingLevel as idempotent.
• Enhance parameter descriptions with format and constraint details. For 'uri' in readResource: 'The resource URI (e.g., file:///path/to/file or https://example.com/resource). Must be a valid URI format.'
• Add error recovery guidance. Instead of 'Failed to list tools: ...', return 'Failed to list tools: connection timeout. Verify the MCP server is running and accessible, then retry.' or 'Tool not found. Call listTools() to see available tools.'

Inferred effective spec: <=2025-11-25.

• Relies on Sampling (deprecated) - integrate directly with the LLM provider API