When developing with an AI agent like Claude Desktop or Claude Code, it is helpful for the agent to be able to test the code you are developing, as you go. This MCP server lets Claude interact with your MCP as a real client, run the tools, execute prompts using an optional LLM configuration, and gain insight into how well the MCP complies with the standard you are going for. You don't have to add your new MCP to the Claude configuration. This one serves as a bridge.

I developed this to solve a real problem I was having. When I would have issues with an MCP deployed to my test cluster, Claude would fail to use the proper tools for it (like cMCP, etc.), and assume the client side was just set up wrong and start rewiring it for Websocket or REST. With this MCP in place, that has stopped, and now I can have Claude use all the tools, prompts and resources for an MCP and then with that in context, help me create an agent that uses the MCP.