MCP Inspector exposes only 2 meta-tools (listTools, callTool) for introspection, not domain-specific functionality. Both tools have minimal descriptions (under 100 chars) and lack comprehensive parameter documentation. listTools has an empty input schema ({}), which violates schema completeness. callTool's 'args' parameter is typed as generic 'object' with no property schema, making it impossible for LLMs to understand what arguments are valid for any given tool. No output schemas are documented. Error handling is absent, no guidance on recovery or categorization. The tool set is intentionally generic (this is an inspector, not a domain tool), but even generic tools must meet quality baselines.

• Add explicit output schema documentation to listTools: return type should be {tools: Array<{name: string, description: string, inputSchema: JSONSchema}>}. This lets LLMs understand the structure before calling.
• For callTool, document the 'args' parameter with an example and constraints: 'Arguments object matching the target tool's inputSchema. Keys and types depend on the tool, call listTools() first to inspect the schema.'
• Add parameter descriptions to both tools. E.g., for callTool's 'name' param: 'The exact name of the tool to invoke (case-sensitive). Call listTools() to discover available tool names.'
• Implement error classification in callTool responses: distinguish 'tool not found' (suggest listTools), 'invalid arguments' (show schema), and 'execution failed' (show error details). Return structured error objects, not plain strings.
• Document that listTools is a discovery tool and should be called first when the agent is unfamiliar with available tools. Add a note: 'Call this before callTool to inspect tool schemas and understand what arguments each tool accepts.'

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

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