mirror of
https://github.com/langgenius/dify-official-plugins.git
synced 2026-07-21 17:45:23 -04:00
The Ollama model cannot be streamed in MCP #601
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @ssl7828 on GitHub (Aug 28, 2025).
Self Checks
Dify version
1.7.1
Plugin version
0.0.7
Cloud or Self Hosted
Self Hosted (Docker)
Steps to reproduce
When I select Ollama's model in the mcp agent node, here I choose the on-premises Qwen3-32b, and find that the ai output is not streamed, but waits for a large section to be returned in the end, and if I switch to the model of a third-party service provider, I can stream the output.
✔️ Error log
No response
@dosubot[bot] commented on GitHub (Aug 28, 2025):
This is a known limitation: the Ollama plugin does not support streaming output in agent nodes (MCP) in Dify, even in recent versions. Responses are only returned after the full answer is ready, not incrementally streamed. Streaming works with some third-party providers and in other application types like Chatflow, but not with Ollama in agent workflows. This is due to the lack of support for the "stream-tool-call" feature in the Ollama plugin and the way agent nodes handle tool calls and ReAct mode. There is a planned feature to add streaming support for agent nodes in a future Dify release, but it's not available yet for Ollama models. For streaming output in agent nodes, you need to use supported models like Tongyi (Qwen-max) and select Function Calling mode instead of ReAct see details more info.
To reply, just mention @dosu.
How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other
