Show HN: An MCP Server for Testing MCP Servers Using Claude Code (github.com)

mcp-test-mcp is a new testing harness that acts as an MCP (Model Context Protocol) server specifically designed to let AI assistants (notably Claude Desktop/Code) test other MCP servers. It solves the "broken loop" problem where assistants trying to validate MCP implementations can't see tool schemas or execution results and end up breaking working code by converting it to REST/WebSocket patterns. By exposing a suite of testable MCP tools (connect_to_server, list_tools, call_tool, list_resources, list_prompts, execute_prompt_with_llm, etc.), mcp-test-mcp gives assistants the visibility and control to discover, call, and verify remote MCP services end-to-end through the MCP protocol itself. Technically it uses a dual-role architecture: the server role exposes testing tools to Claude, while an embedded MCP client (ConnectionManager) connects to target MCP servers to exercise them. It supports stdio and streamable-http transports (auto-detected), detailed, verifiable outputs (timing, token usage, structured errors), automatic JSON extraction from markdown, and optional LLM integration via any OpenAI-compatible endpoint for full prompt execution. The package targets Python 3.11+, FastMCP v2.12.4+, and ships install options via npx or pip with straightforward integration into Claude configs. For MCP/agent developers this means faster, safer debugging, reliable schema inspection, and reproducible end-to-end tests without forcing protocol rewrites.