v0.0.6

Feat: support MCP SSE params.

• Before: only URL that ends with /sse like "https:/xxx/sse"
• After: Compatible with /sse?key=value

🔨 Build version

dify-plugin-daemon:v0.1.3-local

v0.0.5

Support HTTP Streamable MCP servers.

• Upward compatible with SSE

🤖 UI-TARS-SDK (Experimental)

• UI-TARS is VLM that is trained for GUI operation on your computer.
• Wrapping nutjs library
• Check README and example chatflow for more details

Warning

This feature is only supported via source code deploy
(Dify plugin's local debug mode)

🔨 Build version of .difypkg

Dify v1.4.2
dify-plugin-daemon:v0.1.2-local

v0.0.4

Update python module to latest

The following is simplified requirement.txt

mcp>=1.6.0
dify_plugin==0.1.0

To achieve this, I updated pydantic-settings version of dify plugin SDK.
langgenius/dify-plugin-sdks#80

Note

mcp==1.1.2 was not compatible with FastMCP

Fixed following issues:

#8 , #9
They related to LLM model config AIModelEntity issue which is solved by dify plugin daemon's update.

🔨 Build version of .difypkg

• Dify v1.2.0
• dify plugin SDK v0.1.0 (official release)
• dify plugin daemon v0.0.7

v0.0.3

🎊 Support multi SSE MCP servers

Chatflow for v0.0.3 contains example config for following MCP servers

• everything
• mcp-simple-arxiv (new)
  https://github.com/andybrandt/mcp-simple-arxiv

v0.0.2

🎉Support installing dify-mcp-client plugin via .difypkg and GitHub, with SSE.

Any stdio MCP server can convert SSE MCP server using mcp-proxy
https://github.com/sparfenyuk/mcp-proxy

Note

Multiple server connection with one Agent Strategy node has not supported yet. (Coming soon)

v0.0.1

Feature

• ☑ Tool
• ☑ Resource
• ☑ Prompt
• ☐ Sampling
• ☐ Roots

Limitation

• Each MCP Agent node can connect a stdio MCP server (only single server in config_json)
• No "human-in-the-loop" by default

My environment: Dify version v1.0.0 (Plugin daemon v0.0.3)

Warning

Caution

mcp_client.difypkg dosen't work. (Go to README.md)
Only source code deploy is supported.