Build flows with the LFX MCP server

import Tabs from '@theme/Tabs'; import TabItem from '@theme/TabItem'; import Icon from "@site/src/components/icon";

Use lfx-mcp to connect a coding agent to a running Langflow instance, and then build, validate, and run flows from your terminal.

Your MCP client starts lfx-mcp as a local standard stdio subprocess. The server calls the Langflow REST API, so flows that the agent creates appear in the Langflow UI.

lfx-mcp differs from these other MCP options:

Use Langflow as an MCP server: Expose existing flows as tools that clients can call.
Use Langflow as an MCP client: Call MCP servers from components on the canvas.

For Bob (IBM) and Claude Code, you can also use the guided setup in Langflow under Settings → <Icon name="Terminal" aria-hidden="true" /> Langflow MCP Client. For more information, see Langflow MCP Client for coding agents.

Prerequisites

A running Langflow instance
A Langflow API key
An LLM provider API key available in Langflow, typically as a global variable
uv, or lfx installed with uv pip install lfx
Claude Code, or another MCP client that supports stdio servers

Connect an agent and build a flow

To connect an agent and create a flow on your Langflow server, follow these steps:

Add lfx-mcp to your MCP client.

Use uvx --from lfx lfx-mcp. There is no standalone lfx-mcp package on PyPI; the binary ships inside the lfx package.

To add the server in Claude Code, run:
bash
```
claude mcp add langflow \
  -e LANGFLOW_SERVER_URL=http://localhost:7860 \
  -e LANGFLOW_API_KEY=<YOUR_LANGFLOW_API_KEY> \
  -- uvx --from lfx lfx-mcp
```
To add the server in Claude Desktop or another stdio client, create an MCP server entry that runs lfx-mcp and sets the same environment variables:
<Tabs> <TabItem value="uvx" label="uvx (recommended)" default>
json
```
{
  "mcpServers": {
    "langflow": {
      "command": "uvx",
      "args": ["--from", "lfx", "lfx-mcp"],
      "env": {
        "LANGFLOW_SERVER_URL": "http://localhost:7860",
        "LANGFLOW_API_KEY": "<YOUR_LANGFLOW_API_KEY>"
      }
    }
  }
}
```
</TabItem> <TabItem value="installed" label="lfx installed">
json
```
{
  "mcpServers": {
    "langflow": {
      "command": "lfx-mcp",
      "env": {
        "LANGFLOW_SERVER_URL": "http://localhost:7860",
        "LANGFLOW_API_KEY": "<YOUR_LANGFLOW_API_KEY>"
      }
    }
  }
}
```
</TabItem> </Tabs>
On macOS, Claude Desktop stores this configuration in ~/Library/Application Support/Claude/claude_desktop_config.json.

Variable Description Default
LANGFLOW_SERVER_URL URL of your Langflow instance http://localhost:7860
LANGFLOW_API_KEY Langflow API key None

If you omit LANGFLOW_API_KEY, the agent can call the login tool with your Langflow username and password.
Confirm that the server is registered.
bash
```
claude mcp list
```
The output includes an entry similar to the following:
```
langflow: uvx --from lfx lfx-mcp
```
Start the agent, and then ask it to build a flow.
bash
```
claude
```
Enter a prompt such as the following:
text
```
Create a simple agent chatbot flow in Langflow using OpenAI, validate the flow,
and then run it with the message "What is Langflow?"
```
The agent usually does the following:
1. Discovers component types with search_component_types or describe_component_type
2. Creates the flow in one request with create_flow_from_spec
3. Validates connections with validate_flow
4. Runs the flow with run_flow, and then returns the response
The new flow appears in the Langflow UI at your server URL.

The create_flow_from_spec tool accepts a compact text specification for nodes, edges, and configuration:
text
```
name: My Chatbot
description: A simple chatbot

nodes:
  A: ChatInput
  B: OpenAIModel
  C: ChatOutput

edges:
  A.message -> B.input_value
  B.text_output -> C.input_value

config:
  B.model_name: gpt-4o-mini
```
To confirm type names and input and output names before you write edges, use describe_component_type. Connecting through component_as_tool enables tool mode for agent tool wiring.

The MCP client discovers the full tool list from the server. In addition to building and running flows, agents can do the following:
- List, duplicate, rename, export, and delete flows
- Create flows from starter projects
- Batch several tool calls with $N.field references between steps
- Call notify_done when edits are finished so that the Langflow UI refreshes

Variable	Description	Default
`LANGFLOW_SERVER_URL`	URL of your Langflow instance	`http://localhost:7860`
`LANGFLOW_API_KEY`	Langflow API key	None

Prerequisites

Connect an agent and build a flow

See also