ERPNext MCP服务器

ERPNext MCP Server

AI-powered Model Context Protocol server for ERPNext and Frappe applications

Publish ERPNext and other Frappe-based apps to an AI assistant as MCP tools. Use with Cursor IDE, Claude Desktop, and any other MCP client.

✨ Features

• 🔌 MCP Protocol - Standard protocol for AI tool integration
• 📊 Generic Tools - Works with ANY ERPNext doctype (standard or custom)
• 📈 Advanced Analytics 🆕 - Aggregations (SUM, COUNT, AVG, TOP N) and ERPNext reports
• 🔐 OAuth2 Authentication - Standard OAuth2 security with token caching
• 🚀 Production Ready - Built with Go for performance

🚀 Quick Start

Option 1: Automated Install (Recommended for MCP)

# Install the STDIO binary for MCP clients (Cursor, Claude Desktop)
curl -fsSL https://raw.githubusercontent.com/vyogotech/frappe-mcp-server/main/install.sh | bash

This installs the MCP server binary to ~/.local/bin/frappe-mcp-server-stdio

Option 2: Manual Install

# 1. Download pre-built binary from releases
# Visit: https://github.com/vyogotech/frappe-mcp-server/releases/latest

# 2. Or build from source
git clone https://github.com/vyogotech/frappe-mcp-server
cd frappe-mcp-server
make build-stdio

# 3. Configure
cp config.yaml.example config.yaml
# Edit config.yaml with your Frappe/ERPNext credentials

📖 Documentation

Complete documentation: https://vyogotech.github.io/frappe-mcp-server/

Key guides:

• Quick Start - Get running in 5 minutes
• Authentication - sid cookie, OAuth2, and API key auth
• Auth Quick Start - Set up auth in 5 minutes
• Docker Deployment - Deploy with Docker Compose
• API Reference - Complete API docs

💡 Usage Examples

Cursor IDE

@erpnext List all open projects
@erpnext Show me customer ABC-CORP
@erpnext What are the pending tasks?
@erpnext Show me top 5 customers by revenue  🆕
@erpnext Run Sales Analytics report  🆕

HTTP API

Every path except the health probes is authenticated when auth.require_auth is on, as config.yaml.example ships it: send the caller's OAuth2 bearer token or their Frappe sid cookie. Without one the server answers 401 Unauthorized.

# List the tools this server publishes
curl http://localhost:8080/api/v1/tools

# Call one of them
curl -X POST http://localhost:8080/api/v1/tools/get_document \
  -H "Content-Type: application/json" \
  -d '{"params": {"doctype": "Project", "name": "PROJ-0001"}}'

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "erpnext": {
      "command": "/path/to/bin/frappe-mcp-server-stdio",
      "args": ["--config", "/path/to/config.yaml"]
    }
  }
}

🏗️ Architecture

┌─────────────────┐
│   AI Clients    │
│ Cursor, Claude  │
└────────┬────────┘
         │ MCP Protocol
         ↓
┌─────────────────┐
│  ERPNext MCP    │
│     Server      │
└────────┬────────┘
         │ REST API
         ↓
┌─────────────────┐
│    ERPNext      │
│   (Frappe API)  │
└─────────────────┘

🛠️ Prerequisites

• Go 1.25+ — go.mod names the toolchain the build uses
• ERPNext instance (local or remote)

📋 Available Tools

Both binaries register the same catalogue, so tools/list answers the same over stdio and over HTTP. search_knowledge_base is registered only where the rag app it reads is installed (tools.knowledge_base). The six legacy names are still callable and carry no description, so a model is never told to reach for them. go test ./internal/tools -run Readme fails when the table below and the catalogue drift apart, and prints the table to paste back in.

The REST listing at GET /api/v1/tools publishes only the described tools; POST /mcp publishes all of them.

🤝 Contributing

Contributions welcome! See Development Guide.

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests
5. Submit a pull request

📄 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

• ERPNext - Open source ERP
• Model Context Protocol - AI integration standard
• Cursor - AI-powered IDE

📞 Support

• Documentation: GitHub Pages
• Issues: GitHub Issues
• Discussions: GitHub Discussions

Made with ❤️ by the community