Prerequisites
- Flowise (self-hosted) — flowiseai.com
- Node.js 18+ on the Flowise server
- Agent MCP Studio at agentmcp.studio
Step-by-Step Setup
Start bridge.js on the Flowise server
curl -O https://agentmcp.studio/bridge.js && npm install ws node bridge.js wss://agentmcp.studio/api/relay/YOUR-UUID
Open the studio → Settings → MCP Relay Bridge → click Connect. Bridge chip shows 🟢 Live.
-
Open Flowise and create a chatflow
In Flowise, create a new Chatflow or open an existing agent flow. From the node panel, drag in:
- An OpenAI Tool Agent or Tool Agent node
- An MCP Tool node from the Tools category
-
Configure the MCP Tool node
Click the MCP Tool node and set:
- Server Type: stdio
- Command:
node - Arguments:
/path/to/bridge.js wss://agentmcp.studio/api/relay/YOUR-UUID
Flowise will automatically detect all available tools from Agent MCP Studio and display them in the node. Connect the MCP Tool node to the Tool Agent's Tools input.
Save and test
Click Save, then use the built-in chat to test. Ask the agent to use one of your tools — Flowise will call bridge.js, relay the request to your browser, and return the result.
Frequently Asked Questions
Yes — Flowise has an MCP Tool node that connects to external MCP servers. Drag it into your chatflow, configure it with your relay URL and bridge.js details, and your Agent MCP Studio tools appear as callable nodes.
No — Flowise is a no-code/low-code tool. You configure the MCP Tool node via Flowise's GUI. The only non-GUI step is placing bridge.js on your Flowise server and pointing the node at it.
You need bridge.js accessible from wherever Flowise runs. For Flowise Cloud, run bridge.js on a machine that can reach the relay, or self-host Flowise on your own server where bridge.js is co-located.
Yes — Flowise lets you combine retrieval (vector store nodes) with tool calling (MCP nodes) in the same chatflow. This enables RAG + custom tool workflows entirely through Flowise's visual interface.
Any model in Flowise that supports function calling — OpenAI GPT-4o, Claude, Gemini, and select local models via Ollama. Configure the LLM node to a function-calling capable model for MCP tools to be invoked.