SkillFoxx

FlowX MCP

An MCP server and Python SDK that build, run and refine AI agent workflows from a natural-language description.

  • 202 ★ on the repository
  • Last commit
  • Apache-2.0
MCP server

Broad access

Why this level

  • The server generates and executes node Python code on your machine and starts a FastAPI backend.
  • The LLM key and generated code sit in a workspace on disk.
All reasons and checks

aiprobuilder/flowxAuthor:AIpRoBuilder

Install

Manual install

python -m pip install -e .

Run in the cloned repository. Then register `flowx-mcp` in the MCP client with an absolute path.

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • The server generates and executes node Python code on your machine and starts a FastAPI backend.
  • The LLM key and generated code sit in a workspace on disk.

This is third-party code. Review the repository files before installing.

What it does

FlowX takes a process description and generates a workflow graph, audited and tested Python nodes and a FastAPI backend. Each business node goes through a chain of generation, audit, test and bounded repair, and a run returns a `base_url` you can connect to. Control goes through MCP (local stdio), the Python SDK or a non-streaming A2A adapter. MCP offers step-by-step `run_step`, session reset with `reset_session` and workflow evolution: refine the whole workflow, a single node or let the engine decide. For VS Code and Hermes Agent there is a separate `flowx-workflows` skill that guides an agent through building, inspecting, repairing and running.

Who it is for. Agent-system developers who want to assemble workflows from a description and drive them from an MCP client.

Good fit when

  • A process is described in words and you need a working graph with node code and a backend.
  • One node failed its test and must be repaired with user guidance.
  • A finished workflow must run step by step and reset its session between runs.
  • A running workflow needs a single node refined.

Not a fit when

  • You need a ready no-code visual builder: FlowX generates Python.
  • Generated code cannot be executed on the machine: nodes run locally.
  • You need streaming over A2A: the adapter does not support it.

Example request

Build a FlowX workflow that accepts a request, validates the data and sends a notification, then run it step by step.

MCP

Transport
stdio
Authentication
API key

Environment variables

FLOWX_LLM_PROVIDER required
LLM provider, for example deepseek.
FLOWX_LLM_MODEL required
LLM model used to generate nodes.
FLOWX_LLM_API_KEY required, secret
API key of the chosen LLM provider.
FLOWX_DEFAULT_WORKSPACE required
Absolute path to the workspaces directory.

Limitations

An LLM key (`FLOWX_LLM_API_KEY`) is required, and provider and model are set by variables, with DeepSeek in the example. Installation runs from source: `python -m pip install -e .`, and the `flowx-mcp` executable must be given by absolute path in the client config because MCP hosts may not inherit `PATH`. Nodes execute generated Python on your machine, so keep keys per client rules and do not commit them. The agent skill installs separately and a plain `pip install` does not copy it. The last commit dates to November 2026.

How to disable. Remove the `flowx` entry from the MCP client config and delete the Python environment with the package and the `flowx_workspaces` directory.

FAQ

How do I install FlowX?

Clone the repository and run `python -m pip install -e .` in a virtual environment, or use `poetry install`. The `flowx-mcp` executable appears in the environment.

How do I connect it to an MCP client?

Add a `flowx` entry with the absolute path to `flowx-mcp` and the variables `FLOWX_LLM_PROVIDER`, `FLOWX_LLM_MODEL`, `FLOWX_LLM_API_KEY` and `FLOWX_DEFAULT_WORKSPACE`.

How do I run a workflow step by step?

With the `run_step` tool, and use `reset_session` between runs.

How do I refine a finished workflow?

Through evolution: it can change the whole workflow, one node or pick the target itself.

Is there an agent skill?

Yes, `flowx-workflows`. For Hermes Agent install it with `hermes skills install AIpRoBuilder/FlowX/skills/flowx-workflows`.

Discussion

No questions or reviews yet.

MCP servers

Flow-Like MCP

Flow-Like

Flow-Like MCP nodes expose visual workflow functions to an agent over HTTP, with owner-selected tools and access

MCP serverExtended access980

Workflow manager CLI: scripts, secrets and automation of many projects behind one interface, plus an MCP server for agents.

CLIBroad access139
Official

The MCP server of the Loonflow platform: an agent reads the ticket and approval queue and acts on tickets

MCP serverBroad access2.1K
Official

apra-fleet runs a fleet of AI agents across machines and models through MCP and drives multi-hour workflows

MCP serverBroad access106

A Python engine that runs browser and API work as steps, traces each one and replays from the failed step. It has a CLI and an MCP server with 482 modules

MCP serverBroad access483

README in short

The README explains the describe, generate, inspect, run, evolve, reuse cycle, lists news on the node engine and runtime, and walks through install, environment setup and MCP client registration.

README badge

Report a problem