MCP tool

MCP - the Model Context Protocol - is the open standard for connecting an LLM to external tools and data sources: a filesystem, a database, a search engine, a company's internal API, whatever. Instead of every app writing its own bespoke integration for every tool, an MCP server exposes a standard interface and any MCP-aware client can talk to it. This node is comfyui_LLM_party's client side of that: connect it to an MCP server and every tool that server exposes becomes something the pack's LLM nodes can call, without you writing a custom tool node per capability.

What makes this one different from the pack's other tool nodes (Google Search Tool, Embeddings Tool) is that it's a gateway rather than a single fixed capability. Google Search Tool always does Google search. This node's actual behavior is entirely defined by whatever MCP server you point it at - connect it to a filesystem server and the LLM gets file read/write tools; connect it to a different server and it gets a completely different tool set. On the node itself there's really only one control, is_enable, because the interesting configuration doesn't live on the node - it lives in mcp_config.json in the pack's project folder, where you list which MCP server(s) to connect to and how. The default configuration points at the "Everything" server, which the Model Context Protocol project ships specifically as a reference/test server for verifying an MCP client actually works, before you swap in something you'd use for real. The output is the same tool string pattern used across the pack's other tool nodes, meant for the tools input on an LLM node.

If you want to see the range of what's possible here, the modelcontextprotocol/servers repository is a good reference - it's a growing catalog of ready-made MCP servers for common integrations, and pointing mcp_config.json at one of those is generally faster than writing your own from scratch.

Installing it means installing the pack: search "comfyui_LLM_party" in ComfyUI Manager, or git clone https://github.com/heshengtao/comfyui_LLM_party into custom_nodes and restart, then pip install -r requirements.txt from inside the pack folder using ComfyUI's own Python (portable installs: python_embeded\python.exe -m pip install -r requirements.txt, not your system pip).

Where this actually gets people: the node has almost no configuration surface itself, so when it's not working, the problem is nearly always in mcp_config.json, not on the ComfyUI canvas - a malformed server entry, a server that isn't actually running, or a server your machine can't reach. Start with the default "Everything" server before wiring in a real one, precisely because it exists to isolate whether the connection mechanism itself works before you add a real integration's complexity on top. And the same rule that applies to the pack's other tool nodes applies here too: the LLM needs to actually know a tool exists and when to reach for it, which usually means your system prompt should say something about it - a model with no hint that tools are available will mostly just answer from its own knowledge and never touch the MCP server at all.