MCP Server Integration

Overview

The AutoMFlows MCP (Model Context Protocol) server enables AI agents like Claude, Cursor, and other MCP-compatible tools to interact with AutoMFlows. AI agents can generate workflows from natural language, execute them, analyze errors, and automatically fix issues.

AI-Powered Creation

Generate workflows from natural language descriptions

Automatic Execution

Execute workflows directly from AI agents

Error Analysis

Analyze execution errors and get suggestions

Auto-Fix

Automatically fix workflows based on error analysis

Features

Tools

The MCP server provides these executable tools:

Tool	Description
`create_workflow`	Generate workflows from natural language descriptions
`execute_workflow`	Execute a workflow on the AutoMFlows backend
`get_execution_status`	Get execution status and results
`analyze_workflow_errors`	Analyze errors and provide suggestions
`fix_workflow`	Automatically fix workflows based on error analysis
`validate_workflow`	Validate workflow structure before execution

Resources

The MCP server exposes these read-only resources:

Resource	URI	Description
Workflow Examples	`automflows://workflow-examples`	List of example workflows
Node Documentation	`automflows://node-documentation`	Node type documentation
Project Context	`automflows://project-context`	Project structure and conventions

Prerequisites

Before setting up the MCP server, ensure you have:

Node.js 18+ and npm 9+
AutoMFlows backend running (for workflow execution)
MCP-compatible client (Cursor, Claude Desktop, etc.)

Installation

Navigate to MCP server directory

cd mcp-server

Install dependencies

npm install

Build the server

npm run build

This compiles TypeScript to JavaScript in the dist/ directory.

Verify build

ls dist/server.js

You should see the compiled server file.

Configuration

The MCP server can be configured using environment variables or a configuration file.

Environment Variables

Core Settings
LLM Provider
OpenAI
Local LLM

Variable	Default	Description
`AUTOMFLOWS_BACKEND_URL`	`http://localhost:3000`	Backend server URL
`AUTOMFLOWS_WORKFLOWS_PATH`	`./tests/workflows/demo`	Path to workflow examples
`AUTOMFLOWS_VERBOSE`	`false`	Enable verbose logging

Variable	Default	Description
`LLM_PROVIDER`	`none`	LLM provider: `openai`, `local`, or `none`

Variable	Default	Description
`OPENAI_API_KEY`	-	Your OpenAI API key (required if using OpenAI)
`OPENAI_MODEL`	`gpt-4`	OpenAI model name

Variable	Default	Description
`LOCAL_LLM_BASE_URL`	`http://localhost:11434`	Local LLM server URL (Ollama)
`LOCAL_LLM_MODEL`	`llama3`	Local LLM model name

Configuration File

Create mcp-server/config.json for persistent configuration:

{
  "backendUrl": "http://localhost:3000",
  "workflowsPath": "./tests/workflows/demo"
}

Cursor IDE Setup

The most common use case is integrating with Cursor IDE for AI-powered workflow development.

Method 1: Using Cursor Settings UI (Recommended)

Open Cursor Settings

macOS: Press Cmd + ,
Windows/Linux: Press Ctrl + ,

Navigate to MCP

Go to Features > MCP or Tools & Integrations > MCP Servers

Add new MCP server

Click ”+ Add New MCP Server” button

Configure server

Fill in the following fields:

Name: automflows (or any name you prefer)
Command: node
Args: ["/absolute/path/to/autoMflows/mcp-server/dist/server.js"]
Environment Variables (optional):
- AUTOMFLOWS_BACKEND_URL: http://localhost:3000
- LLM_PROVIDER: none (or openai/local)
- OPENAI_API_KEY: Your API key (if using OpenAI)

Save and restart

Click Save or Apply, then restart Cursor IDE

Replace /absolute/path/to/autoMflows/mcp-server/dist/server.js with your actual absolute path!Example paths:

macOS: /Users/username/projects/autoMflows/mcp-server/dist/server.js
Linux: /home/username/projects/autoMflows/mcp-server/dist/server.js
Windows: C:\\Users\\username\\projects\\autoMflows\\mcp-server\\dist\\server.js

Method 2: Manual Configuration File

Edit the MCP configuration file directly:

macOS/Linux
Windows

Project-specific (Recommended):

# In your project root
mkdir -p .cursor
nano .cursor/mcp.json

Global:

mkdir -p ~/.cursor/config
nano ~/.cursor/config/mcp.json

Project-specific (Recommended):

# In your project root
New-Item -ItemType Directory -Force -Path .cursor
notepad .cursor\mcp.json

Global:

notepad %USERPROFILE%\.cursor\config\mcp.json

Add the following configuration:

{
  "mcpServers": {
    "automflows": {
      "command": "node",
      "args": ["/absolute/path/to/autoMflows/mcp-server/dist/server.js"]
    }
  }
}

Getting Your Absolute Path

macOS/Linux
Windows (PowerShell)

cd mcp-server
pwd
# Output: /Users/username/projects/autoMflows/mcp-server
# Full path: /Users/username/projects/autoMflows/mcp-server/dist/server.js

cd mcp-server
(Get-Location).Path
# Output: C:\Users\username\projects\autoMflows\mcp-server
# Full path: C:\Users\username\projects\autoMflows\mcp-server\dist\server.js

Verifying the Connection

Restart Cursor IDE

Fully restart Cursor after configuration changes

Check MCP status

Open Command Palette (Cmd+Shift+P / Ctrl+Shift+P) and type “MCP”Look for “MCP: List Servers” or similar commands

Test in Composer

Open Composer (Cmd+I / Ctrl+I) and ask:

What MCP resources are available?

The agent should list AutoMFlows resources and tools

Usage Examples

Creating a Workflow

Ask the AI agent in Cursor Composer:

Create a workflow that:
Opens example.com
Fills in a login form
Clicks the submit button
Takes a screenshot of the result

The MCP server will generate a workflow JSON that the backend can execute.

Executing and Monitoring

// Create a workflow
const workflow = await createWorkflow({
  userRequest: "Log into example.com and fill a form",
  useCase: "User login and form submission"
});

// Validate it
const validation = validateWorkflow({ workflow });

// Execute it
const result = await executeWorkflow({ workflow });

// Monitor execution
const status = await getExecutionStatus({
  executionId: result.executionId,
  pollUntilComplete: true
});

Error Handling and Auto-Fix

// If execution fails
if (status.status === 'error') {
  // Analyze errors
  const analysis = analyzeWorkflowErrors({
    workflow,
    errorMessage: status.error!
  });
  
  // Automatically fix workflow
  const fixed = await fixWorkflow({
    workflow,
    errorAnalysis: analysis,
    errorMessage: status.error!
  });
  
  // Re-execute fixed workflow
  await executeWorkflow({ workflow: fixed });
}

Using Resources

In Cursor Composer, ask:

Show me workflow examples

Development

Running in Development Mode

# Watch mode with auto-rebuild
npm run dev

# Type check
npm run type-check

# Build for production
npm run build

Testing the Server

The server communicates via stdio, so it’s typically run by an MCP client. For manual testing:

# Start the server (it will wait for stdio input)
node dist/server.js

Then send MCP protocol messages via stdin.

Architecture

The MCP server is built with:

Resources: Read-only data provided to AI agents
- Workflow examples from tests/workflows/demo
- Node type documentation
- Project context and conventions
Tools: Executable functions for AI agents
- Workflow creation (with optional LLM enhancement)
- Workflow execution via HTTP/WebSocket
- Error analysis and automatic fixing
LLM Integration: Optional enhancement
- OpenAI for intelligent workflow generation
- Local LLM (Ollama) for privacy-focused setups
- Rule-based fallback when no LLM configured
Backend Integration
- HTTP client for REST API calls
- WebSocket client for real-time execution updates

Troubleshooting

Server Not Found

Checklist:

Ensure the path is absolute (starts with / on Unix, C:\ on Windows)
Verify you’ve run npm run build in the mcp-server directory
Check the file exists: ls dist/server.js (Unix) or dir dist\server.js (Windows)

Node Not Found

# Check Node version (should be 18+)
node --version

# Check Node is in PATH
which node  # Unix
where node  # Windows

If needed, use the full path to Node:

Unix: "/usr/local/bin/node"
Windows: "C:\\Program Files\\nodejs\\node.exe"

Connection Issues

Common causes:

Backend not running: Start with npm run dev:backend
Wrong port: Check AUTOMFLOWS_BACKEND_URL matches your backend port
Firewall blocking connection

Check Cursor logs:

macOS: ~/Library/Logs/Cursor/
Windows: %APPDATA%\Cursor\logs\
Linux: ~/.config/Cursor/logs/

MCP Server Not Appearing

Restart Cursor completely after configuration changes
Validate JSON syntax (no trailing commas)
Check file is in correct location (.cursor/mcp.json or global config)
Review Cursor logs for error messages

Build Errors

# Clean and rebuild
rm -rf dist node_modules
npm install
npm run build

Advanced Configuration

Custom Workflow Path

Point to your own workflow examples:

{
  "env": {
    "AUTOMFLOWS_WORKFLOWS_PATH": "/path/to/my/workflows"
  }
}

Verbose Logging

Enable detailed logging for debugging:

{
  "env": {
    "AUTOMFLOWS_VERBOSE": "true"
  }
}

Multiple Backend Instances

Configure different backends for different projects:

.cursor/mcp.json

{
  "mcpServers": {
    "automflows-dev": {
      "command": "node",
      "args": ["/path/to/mcp-server/dist/server.js"],
      "env": {
        "AUTOMFLOWS_BACKEND_URL": "http://localhost:3003"
      }
    },
    "automflows-staging": {
      "command": "node",
      "args": ["/path/to/mcp-server/dist/server.js"],
      "env": {
        "AUTOMFLOWS_BACKEND_URL": "http://staging.example.com:3000"
      }
    }
  }
}

Next Steps

Building Plugins

Create custom nodes with plugins

Docker Deployment

Deploy AutoMFlows with Docker

API Reference

Explore the REST API

Creating Workflows

Learn how to create workflows

Get Started

Core Concepts

Guides

Advanced

Documentation Index

​Overview

AI-Powered Creation

Automatic Execution

Error Analysis

Auto-Fix

​Features

​Tools

​Resources

​Prerequisites

​Installation

​Configuration

​Environment Variables

​Configuration File

​Cursor IDE Setup

​Method 1: Using Cursor Settings UI (Recommended)

​Method 2: Manual Configuration File

​Getting Your Absolute Path

​Verifying the Connection

​Usage Examples

​Creating a Workflow

​Executing and Monitoring

​Error Handling and Auto-Fix

​Using Resources

​Development

​Running in Development Mode

​Testing the Server

​Architecture

​Troubleshooting

​Server Not Found

​Node Not Found

​Connection Issues

​MCP Server Not Appearing

​Build Errors

​Advanced Configuration

​Custom Workflow Path

​Verbose Logging

​Multiple Backend Instances

​Next Steps

Building Plugins

Docker Deployment

API Reference

Creating Workflows

Overview

Features

Tools

Resources

Prerequisites

Installation

Configuration

Environment Variables

Configuration File

Cursor IDE Setup

Method 1: Using Cursor Settings UI (Recommended)

Method 2: Manual Configuration File

Getting Your Absolute Path

Verifying the Connection

Usage Examples

Creating a Workflow

Executing and Monitoring

Error Handling and Auto-Fix

Using Resources

Development

Running in Development Mode

Testing the Server

Architecture

Troubleshooting

Server Not Found

Node Not Found

Connection Issues

MCP Server Not Appearing

Build Errors

Advanced Configuration

Custom Workflow Path

Verbose Logging

Multiple Backend Instances

Next Steps