If you're running local LLMs through llama.cpp, Ollama, or vLLM and want them to use MCP tools, this is the bridge. It translates between OpenAI-compatible chat completion endpoints and MCP servers, letting any function-calling model access the same tools Claude Desktop uses. The config format is intentionally identical to Claude Desktop's mcpServers block, so you can copy it directly. Built for the qso-graph amateur radio ecosystem but works with any MCP server. Write operations are disabled by default. Useful when you need local inference with tool calling but don't want to rewrite tool integrations or move weights to the cloud.
claude mcp add --transport stdio qso-graph-qsp-mcp uvx qsp-mcp