Model Context Protocol Server

What is Model Context Protocol Server about?

The server acts as a bridge between Dify’s workflow engine and any client that supports the Model Context Protocol (MCP). By exposing Dify workflows as MCP tools, developers can trigger complex AI pipelines without writing custom integration code.

How to use Model Context Protocol Server?

Configure credentials – Provide the Dify base URL and one or more App Secret Keys (SKs) either via environment variables (DIFY_BASE_URL, DIFY_APP_SKS) or a config.yaml file.
Install the runtime – Install uv/uvx (a fast Python package manager) with curl -Ls https://astral.sh/uv/install.sh | sh.
Run the server – Use one of the recommended command patterns:
- uvx (no clone): uvx --from git+https://github.com/YanxingLiu/dify-mcp-server dify_mcp_server
- uv + local clone: uv --directory /path/to/dify-mcp-server run dify_mcp_server
Point clients to the server – Any MCP‑compatible client can now call the Dify tools exposed by this server.

Key features of Model Context Protocol Server

Multi‑workflow support – Accepts a comma‑separated list of Dify App SKs, exposing each as a separate MCP tool.
Dual configuration method – Environment variables for cloud‑native deployments or a simple config.yaml for local testing.
Zero‑clone option – uvx runs the server directly from the Git repository, simplifying CI/CD integration.
Lightweight Python implementation – Minimal dependencies, fast startup via uv.

Use cases of Model Context Protocol Server

Integrating Dify AI assistants into custom chatbots that rely on MCP for tool calls.
Automating business processes where Dify workflows handle data enrichment, summarisation, or classification.
Rapid prototyping of AI‑augmented features in SaaS products without building bespoke Dify SDK layers.
Cloud‑native deployments where environment variables are injected by platforms such as Vercel, Railway, or Kubernetes.

FAQ from the Model Context Protocol Server

Q: Do I need to clone the repository to run the server? A: No. Using uvx you can run it directly from Git. Cloning is only required if you prefer a local development setup.

Q: How many Dify workflows can I expose? A: Any number; just list their App SKs in DIFY_APP_SKS (comma‑separated) or add them to the dify_app_sks array in config.yaml.

Q: What if I want to change the base URL after the server starts? A: Update the environment variable or the YAML file and restart the server; the new value will be read on startup.

Q: Is there a Docker image available? A: The README does not mention an official Docker image, but you can containerise the server by installing uv inside a base Python image and running the same command.

Q: Which Python version is required? A: The project follows standard Python 3.x conventions; uv will resolve compatible versions automatically.

Model Context Protocol (MCP) Server for dify workflows

A simple implementation of an MCP server for using dify. It achieves the invocation of the Dify workflow by calling the tools of MCP.

📰 News

[2025/4/15] zNow supports directly using environment variables to pass base_url and app_sks, making it more convenient to use with cloud-hosted platforms.

🔨Installation

The server can be installed via Smithery or manually.

Step1: prepare config.yaml or enviroments

You can configure the server using either environment variables or a config.yaml file.

Method 1: Using Environment Variables (Recommended for Cloud Platforms)

Set the following environment variables:

export DIFY_BASE_URL="https://cloud.dify.ai/v1"
export DIFY_APP_SKS="app-sk1,app-sk2" # Comma-separated list of your Dify App SKs

DIFY_BASE_URL: The base URL for your Dify API.
DIFY_APP_SKS: A comma-separated list of your Dify App Secret Keys (SKs). Each SK typically corresponds to a different Dify workflow you want to make available via MCP.

Method 2: Using `config.yaml`

Create a config.yaml file to store your Dify base URL and App SKs.

Example config.yaml:

dify_base_url: "https://cloud.dify.ai/v1"
dify_app_sks:
  - "app-sk1" # SK for workflow 1
  - "app-sk2" # SK for workflow 2
  # Add more SKs as needed

dify_base_url: The base URL for your Dify API.
dify_app_sks: A list of your Dify App Secret Keys (SKs). Each SK typically corresponds to a different Dify workflow.

You can create this file quickly using the following command (adjust the path and values as needed):

# Create a directory if it doesn't exist
mkdir -p ~/.config/dify-mcp-server

# Create the config file
cat > ~/.config/dify-mcp-server/config.yaml <<EOF
dify_base_url: "https://cloud.dify.ai/v1"
dify_app_sks:
  - "app-your-sk-1"
  - "app-your-sk-2"
EOF

echo "Configuration file created at ~/.config/dify-mcp-server/config.yaml"

When running the server (as shown in Step 2), you will need to provide the path to this config.yaml file via the CONFIG_PATH environment variable if you choose this method.

Step2: Installation on your client

❓ If you haven't installed uv or uvx yet, you can do it quickly with the following command:

curl -Ls https://astral.sh/uv/install.sh | sh

✅ Method 1: Use uvx (no need to clone code, recommended)

{
"mcpServers": {
  "dify-mcp-server": {
    "command": "uvx",
      "args": [
        "--from","git+https://github.com/YanxingLiu/dify-mcp-server","dify_mcp_server"
      ],
    "env": {
       "DIFY_BASE_URL": "https://cloud.dify.ai/v1",
       "DIFY_APP_SKS": "app-sk1,app-sk2",
    }
  }
}
}

{
"mcpServers": {
  "dify-mcp-server": {
    "command": "uvx",
      "args": [
        "--from","git+https://github.com/YanxingLiu/dify-mcp-server","dify_mcp_server"
      ],
    "env": {
       "CONFIG_PATH": "/Users/lyx/Downloads/config.yaml"
    }
  }
}
}

✅ Method 2: Use uv (local clone + uv start)

You can also run the dify mcp server manually in your clients. The config of client should like the following format:

{
"mcpServers": {
  "mcp-server-rag-web-browser": {
    "command": "uv",
      "args": [
        "--directory", "${DIFY_MCP_SERVER_PATH}",
        "run", "dify_mcp_server"
      ],
    "env": {
       "CONFIG_PATH": "$CONFIG_PATH"
    }
  }
}
}

{
"mcpServers": {
  "mcp-server-rag-web-browser": {
    "command": "uv",
      "args": [
        "--directory", "${DIFY_MCP_SERVER_PATH}",
        "run", "dify_mcp_server"
      ],
    "env": {
       "CONFIG_PATH": "$CONFIG_PATH"
    }
  }
}
}

Example config:

{
"mcpServers": {
  "dify-mcp-server": {
    "command": "uv",
      "args": [
        "--directory", "/Users/lyx/Downloads/dify-mcp-server",
        "run", "dify_mcp_server"
      ],
    "env": {
       "DIFY_BASE_URL": "https://cloud.dify.ai/v1",
       "DIFY_APP_SKS": "app-sk1,app-sk2",
    }
  }
}
}

Enjoy it

At last, you can use dify tools in any client who supports mcp.

Model Context Protocol Server