dify-mcp-client

MCP Client as Agent Strategy Plugin with Computer Using Agent (UI-TARS-SDK) support.

!showcase1

How it works

Each MCP client (ReAct Agent) node can connect MCP servers.

1. Tool, Resource, Prompt lists are converted into Dify Tools.
2. Your selected LLM can see their name, description, argument type
3. The LLM calls Tools based on the ReAct loop (Reason → Act → Observe).

✅ What I did

• Copied ReAct.py and renamed file as mcpReAct.py
• Added config_json GUI input field by editing mcpReAct.yaml and class mcpReActParams()

in mcpReAct.py, I added

• New 12 functions for MCP
• __init__() for initializing AsyncExitStack and event loop
• Some codes in _handle_invoke_action() for MCP
• MCP setup and cleanup in _invoke()

🔄 Update history

• Add SSE MCP client (v0.0.2)
• Support multi SSE servers (v0.0.3)
• Update python module and simplify its dependency (v0.0.4)
  • mcp(v1.1.2→v1.6.0+)
  • dify_plugin(0.0.1b72→v0.1.0)
• Add UI-TARS SDK integration for GUI automation capabilities (v0.0.5)
• Support Streamable HTTP MCP client
• Feat SSE param: /sse?key=value (v0.0.6)

🤖 UI-TARS Integration

This plugin includes UI-TARS SDK integration for GUI automation capabilities.

Key Features

• On-demand GUI automation: UI-TARS is called only when needed, reducing token consumption
• Life-time control: Set maximum loop count per task to prevent runaway automation

Known Limitations

• Single Monitor Support: UI-TARS currently recognizes the primary monitor only. Multi-monitor setups are not supported.
• Mac Retina Display Issue: On macOS with Retina displays, UI-TARS requires the display resolution to be set to "Default" instead of the highest quality setting. Otherwise wrong (w,h) point is clicked. https://github.com/bytedance/UI-TARS-desktop/issues/591

Life-time Parameter

The life_time parameter controls the maximum number of GUI actions UI-TARS can perform:

• Default: 10 iterations
• User-configurable maximum via ui_tars_max_life_time_count
• Your selected LLM can dynamically adjust within the user-defined limit based on task complexity

🐳 Docker Deployment with Pre-built Node.js

Building the Docker Image

docker build -t dify-mcp-client:latest .

# In your docker-compose.yml
services:
  plugin-daemon:
    image: memedayo/dify-plugin-daemon:latest  # with Pre-built Node.js
    # ... rest of configuration

UI-TARS Configuration

For detailed UI-TARS setup, refer to the UI-TARS Desktop deployment guide.

The plugin automatically configures UI-TARS as a tool within the ReAct loop. You need to provide:

• Hugging Face Inference Endpoint URL
• API Key like (hf_xxxxx)
• (Optional) Adjust ui_tars_max_life_time_count in agent parameters

⚠️ Caution and Limitation

How to use this plugin

🛜Install the plugin from GitHub

• Enter the following GitHub repository name

https://github.com/3dify-project/dify-mcp-client/

• Dify > PLUGINS > + Install plugin > INSTALL FROM > GitHub !difyUI1

⬇️Install the plugin from .difypkg file

• Go to Releases https://github.com/3dify-project/dify-mcp-client/releases
• Select suitable version of .difypkg
• Dify > PLUGINS > + Install plugin > INSTALL FROM > Local Package File !difyUI2

How to handle errors when installing plugins?

Issue: If you encounter the error message: plugin verification has been enabled, and the plugin you want to install has a bad signature, how to handle the issue?
Solution: Open /docker/.env and change from true to false:

FORCE_VERIFYING_SIGNATURE=false

Run the following commands to restart the Dify service:

cd docker
docker compose down
docker compose up -d

Once this field is added, the Dify platform will allow the installation of all plugins that are not listed (and thus not verified) in the Dify Marketplace.

Where does this plugin show up?

• It takes few minutes to install
• Once installed, you can use it any workflows as Agent node
• Select "mcpReAct" strategy (otherwise no MCP) !asAgentStrategiesNode

Config

MCP Agent Plugin node require config_json like this to command or URL to connect MCP servers

{
    "mcpServers":{
        "name_of_server1":{
            "url": "http://host.docker.internal:8080/sse"
        },
        "name_of_server2":{
            "url": "http://host.docker.internal:8008/mcp"
        }
    }
}

Chatflow Example

!showcase2

I provide this Dify ChatFlow .yml for testing this plugin.

https://github.com/3dify-project/dify-mcp-client/tree/main/test/chatflow

After download DSL(yml) file, import it in Dify and you can test MCP using "Everything MCP server"

https://github.com/modelcontextprotocol/servers/tree/main/src/everything

How to convert stdio MCP server into Stremable HTTP (or SSE)

option1️⃣: Edit MCP server's code

If fastMCP server, change like this

if __name__ == "__main__":
-    mcp.run(transport="stdio")
+    mcp.run(transport="streamable-http")

option2️⃣: via mcp-proxy

\mcp-proxy>uv venv -p 3.12
.venv\Scripts\activate
uv tool install mcp-proxy

which npx

where npx

C:\Program Files\nodejs\npx
C:\Program Files\nodejs\npx.cmd
C:\Users\USER_NAME\AppData\Roaming\npm\npx
C:\Users\USER_NAME\AppData\Roaming\npm\npx.cmd

{
  "mcpServers": {
    "SERVER_NAME": {
       "command": CMD_NAME_OR_PATH 
       "args": {VALUE1, VALUE2}
    }
  }
}

mcp-proxy --sse-port=8080 --pass-environment -- CMD_NAME_OR_PATH --arg1 VALUE1 --arg2 VALUE2 ...

mcp-proxy --sse-port=8080 --pass-environment -- C:\Program Files\nodejs\npx.cmd --arg1 -y --arg2 @modelcontextprotocol/server-everything

pip install mcp-simple-arxiv
mcp-proxy --sse-port=8008 --pass-environment -- C:\Users\USER_NAME\AppData\Local\Programs\Python\Python310\python.exe -m -mcp_simple_arxiv

(mcp_proxy) C:\User\USER_NAME\mcp-proxy>mcp-proxy --sse-port=8080 --pass-environment -- C:\Program Files\nodejs\npx.cmd --arg1 -y --arg2 @modelcontextprotocol/server-everything
DEBUG:root:Starting stdio client and SSE server
DEBUG:asyncio:Using proactor: IocpProactor
DEBUG:mcp.server.lowlevel.server:Initializing server 'example-servers/everything'
DEBUG:mcp.server.sse:SseServerTransport initialized with endpoint: /messages/
INFO:     Started server process [53104]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://127.0.0.1:8080 (Press CTRL+C to quit)

🔨 How to develop and deploy plugin

Official plugin dev guide

https://github.com/3dify-project/dify-mcp-client/blob/main/GUIDE.md

Dify plugin SDK daemon

If your OS is Windows and CPU is Intel or AMD, you need to download the latest dify-plugin-windows-amd64.exe
Choose your OS-compatible verson here:
https://github.com/langgenius/dify-plugin-daemon/releases

1. Rename it as dify.exe for convinence
2. mkdir "C\User\user\.local\bin" (Windows) and register it as system path.
3. Copy dify.exe to under dify-mcp-client/

Reference

https://docs.dify.ai/plugins/quick-start/develop-plugins/initialize-development-tools

dify plugin init

Change directory

cd dify-mcp-client

Install python module

Python3.12+ is compatible. The venv and uv are not necessary, but recommended.

uv venv -p 3.12
.venv\Scripts\activate

Install python modules for plugin development

uv pip install -r requirements.txt

For only UI-TARS-SDK user (after installing Node.js v22 LTS)

npm install

Duplicate env.example and rename one to .env

I changed REMOTE_INSTALL_HOST from debug.dify.ai to localhost (Docker Compose environment) click 🪲bug icon button to see these information

Activate Dify plugin

python -m main

(ctrl+C to stop)

Package into .difypkg

./dify-mcp-client is my default root name

dify plugin package ./ROOT_OF_YOUR_PROJECT

Useful GitHub repositories for developers

Dify Plugin SDKs

https://github.com/langgenius/dify-plugin-sdks

MCP Python SDK

https://github.com/modelcontextprotocol/python-sdk