The support for MCP is a bit poor. We may have to consider using other frameworks.

[bidada-bcr]

What happened?

The call of MCP is very unstable, and the multi-agent collaboration is also a bit poor. When using MCP, it feels like Autogen doesn't know what it should do, especially when it involves multiple MCPs.

Which packages was the bug in?

Python Extensions (autogen-ext)

AutoGen library version.

Python dev (main branch)

Other library version.

No response

Model used

No response

Model provider

None

Other model provider

No response

Python version

None

.NET version

None

Operating system

None

[SongChiYoung]

Thanks for reporting this issue

I’ve been exploring MCP and multi-agent setups as well, and I’m curious to better understand what might be happening here.

• You mentioned using multiple MCPs at the same time — could you clarify how they were structured? (e.g., independent teams vs. nested or sequential composition)
• Did the issue occur during initialization, message passing, or result aggregation?
• When you say “Autogen doesn’t know what it should do,” was there a specific behavior or output that seemed wrong or unexpected?

If you're able to share a minimal example or part of the setup, that would really help in understanding the context.