Overview
What is mcp-server-dify?
mcp-server-dify is a Model Context Protocol server that enables LLMs to interact with Dify AI’s chat completion API. Written in TypeScript, it provides a standardized interface for chat, streaming responses, and conversation context.
How to use mcp-server-dify?
Install via Docker (make docker then docker run -i --rm mcp/dify <endpoint> <key>) or run directly with npx. Configure it in Claude Desktop by adding the endpoint and API key to your claude_desktop_config.json. Invoke its tools, such as the meshi-doko tool for restaurant recommendations, by supplying the required parameters (LOCATION, BUDGET, query, optional conversation_id).
Key features of mcp-server-dify
- Integration with Dify AI chat completion API
- Restaurant recommendation tool (meshi-doko)
- Support for maintaining conversation context
- Streaming response support
- TypeScript implementation with CI, linting, and formatting
- Docker and npx deployment options
Use cases of mcp-server-dify
- Querying Dify AI for restaurant recommendations based on location and budget
- Building conversational agents that maintain multi-turn context
- Enabling LLMs to access Dify AI’s specialized chat capabilities through a standardized protocol
- Rapid prototyping with Docker or direct CLI invocation
FAQ from mcp-server-dify
What are the prerequisites for using mcp-server-dify?
You need a Dify AI API endpoint URL and an API key. Either Docker (to run the container) or Node.js (to use npx) is required.
How do I install and run mcp-server-dify with Docker?
Build the image with make docker, then run: docker run -i --rm mcp/dify <your-dify-api-endpoint> <your-dify-api-key>.
What does the meshi-doko tool do?
It is a restaurant recommendation tool that interfaces with Dify AI. It accepts a location, budget, query, and optional conversation ID to maintain context across calls.
Does mcp-server-dify support streaming?
Yes, streaming response support is listed as a feature.
How should I secure my API credentials?
Keep your API key private, use HTTPS for the endpoint, and never commit the key to version control.
