OPEN SOURCE / LANGFLOW MCP

Your flows. In your AI workspace.

Connect Langflow to Codex, Claude Code and other MCP clients. Explore your flows, run workflows and work with knowledge bases from the tools you already use.

Connect Langflow Explore the repository

CodexClaude CodeClaude DesktopMCP

01 / MCPstdio · HTTP

One connection. Your whole workflow.

“List my flows and check the latest run.”

01Your MCP client

02Langflow MCP servertools/call

03Your Langflow instanceREST API

A connection diagram, not a live execution.

236standard tools

29consolidated tools

MITlicense

1.12.0Langflow API target

01 / Capabilities

From a saved flow to a working result.

The repository exposes Langflow operations as MCP tools. Choose individual tools or a smaller, action-based interface.

Flows & projects

Find, create, update and organize flows. Import, export and manage their versions.

Execution & human input

Run workflows, inspect results and resume runs waiting for a human decision.

Knowledge & memory

Manage knowledge bases, ingestion and memory. Availability depends on your Langflow configuration.

Models & governance

Discover model providers and inspect policies. Administrative changes use your Langflow permissions.

Monitoring

Inspect messages, sessions and execution traces. Check health and readiness.

Agentic & A2A

Use agentic assistance and Agent-to-Agent endpoints when enabled on your Langflow server.

02 / How it works

Bring the flow to the conversation.

01
Connect
Point the server at your Langflow URL and API key.
02
Ask
Your MCP client discovers the available tools and sends a request.
03
Inspect
Read the result or error returned by Langflow before taking the next action.

03 / Install

Start with your own Langflow.

You need Node.js 20+, a running Langflow instance and its API key. Build from source for the capabilities described on this page.

01 / Build the source

git clone https://github.com/nobrainer-tech/langflow-mcp.git
cd langflow-mcp
npm ci
npm run build

02 / Connect your client

Replace the example URL and API key locally. Run the client command from the cloned repository. Never commit your key.

Codex

codex mcp add langflow \
  --env LANGFLOW_BASE_URL=http://localhost:7860 \
  --env LANGFLOW_API_KEY=YOUR_LANGFLOW_API_KEY \
  --env LANGFLOW_CONSOLIDATED_TOOLS=true \
  -- node "$PWD/dist/mcp/index.js"

Claude Code

claude mcp add langflow \
  -e LANGFLOW_BASE_URL=http://localhost:7860 \
  -e LANGFLOW_API_KEY=YOUR_LANGFLOW_API_KEY \
  -e LANGFLOW_CONSOLIDATED_TOOLS=true \
  -- node "$PWD/dist/mcp/index.js"

Running the server elsewhere?

The repository also supports Streamable HTTP at /mcp. Use HTTPS and bearer authentication for a remote deployment; the README contains the setup details.

Read the HTTP setup ↗

04 / FAQ

A few things worth knowing.

Is this Langflow itself?

No. It is an open-source MCP bridge to your existing Langflow instance. You run and configure Langflow separately.

Which clients can connect?

The repository documents Codex, Claude Code and Claude Desktop. Other clients need compatible MCP stdio or Streamable HTTP support. Support for every client has not been tested end to end.

Can it change or delete data?

Yes. Some tools modify flows, projects and administrative settings. Review requested actions and use an API key with appropriate permissions.

Are streaming results shown live?

Several upstream endpoints use SSE. The client currently buffers their text rather than exposing every event incrementally.

Your Langflow. Your tools.

Inspect the source, connect your instance and start with a read-only request.

Explore the repository ↗