Dify as MCP Server
by Yevanchen
This project implements a Dify plugin to expose Dify workflows as a Model Context Protocol (MCP) server for AI clients like Claude. It allows AI clients to discover and use your Dify workflows as tools.
Last updated: N/A
What is Dify as MCP Server?
This is a Dify plugin that acts as a Model Context Protocol (MCP) server, allowing AI clients like Claude Desktop and Cursor to access and utilize Dify workflows as tools.
How to use Dify as MCP Server?
- Configure your Dify application with at least one workflow. 2. In the AI client (e.g., Claude, Cursor), add a new MCP server with the URL
https://your_dify_instance_address/difyapp_as_mcp_server. 3. Save and enable the server in the client settings. 4. Use the tools in the AI client.
Key features of Dify as MCP Server
Supports MCP standard JSON-RPC interface
Automatic tool discovery and registration
Compatible with Claude Desktop/Cursor
Secure SSE connection implementation
Server-side session management
Compliant with the latest MCP Streamable HTTP specification
Use cases of Dify as MCP Server
Providing Dify workflows as tools to Claude and other AI clients
Enabling AI clients to discover and use Dify workflows
Integrating Dify workflows into AI assistants without modifying the workflows
Exposing Dify applications as tools in Cursor Agent
FAQ from Dify as MCP Server
What if I have connection problems?
What if I have connection problems?
Ensure the URL is correct and accessible. Check for firewalls or proxies.
Why are the tools not visible?
Why are the tools not visible?
Ensure the application ID is configured correctly. Check if the workflow is published. Confirm the Dify API key has sufficient permissions.
What if tool execution fails?
What if tool execution fails?
Check the Dify application logs. Confirm the workflow runs normally in Dify.
How does the server handle sessions?
How does the server handle sessions?
The server generates and manages session IDs, supporting a stateless server mode suitable for Dify's plugin environment.
What is the architecture of the plugin?
What is the architecture of the plugin?
The plugin uses two endpoints: a GET endpoint for SSE connections and HTML pages, and a POST endpoint for JSON-RPC requests.