mcp-flowise

MCP server that exposes the chatflows of your local Flowise as tools for free-code (or any MCP client). Standalone and minimalist.

Tools

Simple mode (default):

• list_chatflows() — lists the chatflows (id + name).

• create_prediction(chatflow_id, question) — runs a chatflow.

Dynamic mode (FLOWISE_DYNAMIC=true):

• Registers one tool per chatflow at startup, e.g. flowise_support_bot(question).

Related MCP server: Coze MCP Server For OpenClaw

Requirements

• Python 3.10+

• uv (recommended) or pip

• A running Flowise instance (default http://localhost:3000)

Configuration

Copy .env.example to .env and fill in at least FLOWISE_API_ENDPOINT and, if your Flowise has auth enabled, FLOWISE_API_KEY (Flowise → Settings → API Keys).

Run it manually (for testing)

cd mcp-flowise
uv run mcp-flowise        # starts over stdio; Ctrl+C to quit
# or, without uv:
pip install -e . && mcp-flowise

To inspect it with the MCP Inspector:

uv run mcp dev mcp_flowise/server.py

Connect it to free-code

Add the server to ~/.free-code/agent/mcp.json (or import it with /mcp-import):

{
  "mcpServers": {
    "flowise": {
      "command": "uvx",
      "args": ["--from", "/absolute/path/to/repo/mcp-flowise", "mcp-flowise"],
      "env": {
        "FLOWISE_API_ENDPOINT": "http://localhost:3000",
        "FLOWISE_API_KEY": ""
      }
    }
  }
}

Then in free-code:

/mcp enable flowise
/reload

New servers start disabled, which is why you need /mcp enable and /reload (or a new session) for the tools to show up.

Notes on the Flowise API

• List: GET {endpoint}/api/v1/chatflows

• Predict: POST {endpoint}/api/v1/prediction/{chatflowId} with {"question": "..."}

• Auth: Authorization: Bearer <FLOWISE_API_KEY> header (if enabled).