Labsco
MCP SERVER

Flowise

by andydukes ↗

Bring your Flowise chatflows into an MCP client — each one registered as its own tool, or two static tools if you would rather keep it simple.

Workflow Automation & iPaaS
Summary
Pick the mode before you pick anything else.

LowLevel mode is the interesting one: your chatflows arrive as distinct tools with distinct descriptions, so the agent routes to the right flow instead of you naming it every time. That only works if the descriptions are good, which is what the descriptions variable is for — leave it empty and the agent is choosing between chatflow names. Simple mode is the right call when there is one flow that matters. Whitelists take precedence over blacklists when both are set.

What it is

A Python MCP server that connects to a Flowise instance and exposes your chatflows to an MCP client. It runs in one of two modes, and the choice matters: LowLevel mode registers every chatflow as its own separate tool, so an agent can pick between them; FastMCP mode exposes two static tools instead, which is less to configure and less for the model to reason about.

What you get
  • Every chatflow on your Flowise instance registered as an individual tool, named after the chatflow and described from `FLOWISE_CHATFLOW_DESCRIPTIONS`, so the agent chooses which flow fits the question — the default LowLevel mode
  • Or, with `FLOWISE_SIMPLE_MODE=true`, two tools instead of many: one to list the chatflows and one to run a prediction against a configured chatflow or assistant — `list_chatflows`, `create_prediction`
  • Filtering in either mode, so a busy Flowise instance does not dump everything into your client: whitelist or blacklist by chatflow ID, or by name with a regular expression — `FLOWISE_WHITELIST_ID`, `FLOWISE_BLACKLIST_ID`, `FLOWISE_WHITELIST_NAME_REGEX`, `FLOWISE_BLACKLIST_NAME_REGEX`
  • A documented Windows path that avoids the `--from git+https` problem: clone locally and point `uvx.exe` at the directory
Requirements

A Flowise instance and its API bearer token: `FLOWISE_API_KEY` is required, `FLOWISE_API_ENDPOINT` defaults to `http://localhost:3010`. Python 3.12 or higher and the `uvx` package manager; run straight from the repository with `uvx --from git+https://github.com/andydukes/mcp-flowise mcp-flowise`, or install through Smithery for Claude Desktop. The package is `mcp-flowise` (0.1.0 in pyproject). Setting both `FLOWISE_CHATFLOW_ID` and `FLOWISE_ASSISTANT_ID` makes the server refuse to start — pick one.

Setup effort

One command plus a key — uvx --from git+https://github.com/andydukes/mcp-flowise mcp-flowise, then supply credentials