FastMCP import fails after httpx≥0.28 upgrade (TransportError removed)

[ghbaud]

Title: FastMCP import fails after httpx≥0.28 upgrade (TransportError removed)

• MCP version: 1.20.0 (also happens on 1.18.x)
• Python: 3.12.10 on Windows 11
• Install method: local venv (pip install mcp[cli])

Steps to reproduce

1. Create a fresh venv.
2. pip install mcp[cli]>=1.2.0.
3. Run python -m mcp.dev path/to/server.py or simply from mcp.server.fastmcp import FastMCP.

Observed error

from mcp.client.sse import sse_client
  File ".../site-packages/mcp/client/sse.py", line 10, in <module>
    from httpx_sse import aconnect_sse
  File ".../site-packages/httpx_sse/_exceptions.py", line 4, in <module>
    class SSEError(httpx.TransportError):
AttributeError: module 'httpx' has no attribute 'TransportError'

Notes

• Resolver now picks httpx 0.28.1 + httpx-sse 0.4.3.
• httpx ≥0.28 removed/renamed TransportError; httpx_sse still subclasses it.
• Pinning httpx==0.26.0 avoids the crash but conflicts with MCP’s httpx>=0.27.1.
• Temporary workaround is editing .venv/Lib/site-packages/httpx_sse/_exceptions.py to fall back to httpx.HTTPError; that isn’t sustainable.

Request
Please pin MCP’s dependency stack to a compatible combination (e.g. httpx<0.28) or bump the required httpx-sse version so it handles the API change. If an official workaround already exists, documenting it would help downstream users. Thanks!

[maxisbey]

I can't seem to reproduce this myself on my Mac.

[17:15] ~/code/temp $ uv init --python 3.12
Initialized project `temp`
[17:15] ~/code/temp $ uv add mcp\[cli\]==1.2.0 httpx==0.28.1 httpx-sse==0.4.3
Using CPython 3.12.11 interpreter at: /xxx/python3.12
Creating virtual environment at: .venv
Resolved 28 packages in 1.03s
Prepared 26 packages in 1ms
Installed 26 packages in 22ms
 + annotated-types==0.7.0
 + anyio==4.11.0
 + certifi==2025.10.5
 + click==8.3.0
 + h11==0.16.0
 + httpcore==1.0.9
 + httpx==0.28.1
 + httpx-sse==0.4.3
 + idna==3.11
 + markdown-it-py==4.0.0
 + mcp==1.2.0
 + mdurl==0.1.2
 + pydantic==2.12.3
 + pydantic-core==2.41.4
 + pydantic-settings==2.11.0
 + pygments==2.19.2
 + python-dotenv==1.2.1
 + rich==14.2.0
 + shellingham==1.5.4
 + sniffio==1.3.1
 + sse-starlette==3.0.2
 + starlette==0.48.0
 + typer==0.20.0
 + typing-extensions==4.15.0
 + typing-inspection==0.4.2
 + uvicorn==0.38.0
[17:15] ~/code/temp $ uv run python -c "from mcp.server.fastmcp import FastMCP;print('success')"
success
[17:15] ~/code/temp $

Same when running an actual test script, I don't run into any exceptions.

This could potentially be a Windows specific error?

[ghbaud]

It's a good bet that it's Windows specific. Not being able to reproduce it on a Mac may give you some information, but not enough to fix it. I hope you don't dismiss it as an issue because of that. Have pity on us poor Windows developers!

[maxisbey]

haha no of course, I'll just have to look into it when I get my Windows dev environment set up (hopefully soon)

[chuxinyuan]

great bug for first step.

[chuxinyuan]

It's a good bet that it's Windows specific. Not being able to reproduce it on a Mac may give you some information, but not enough to fix it. I hope you don't dismiss it as an issue because of that. Have pity on us poor Windows developers!

The sample problem on Arch Linux

[Azzybana]

Checking in from Windows, staring at this very issue now. httpx made an API breaking change.

[Kludex]

I can't reproduce it.

I need an MRE.