Doesn't work well on windows

[spivak321c]

[Pringled]

Hey @spivak321c, can you share your environment maybe? I have not seen this error before, did you install the optional MCP dependency group for Semble as well?

[AstroDogeDX]

I think the issue here is that resource is a UNIX exclusive package and not available for Windows, meaning we can't use Semble on a pure Windows environment, which is a shame.

[AstroDogeDX]

Little bit of research suggests that psutil would be closest cross-platform alternative, but I haven't peeped the source to understand how resource is being used; so I'm not sure how viable it'd be as a replacement or platform alternative with a wrapper.

[Pringled]

@AstroDogeDX if that's the case we definitely want to fix that. However, I think that the issue is actually that the MCP dependencies are not installed. Up to 2 weeks ago, this resource warning appeared when importing BM25s (which Semble does), see https://github.com/xhluca/bm25s/releases/tag/0.3.5. I don't have a Windows environment to test this on though (but we do run Windows in CI), are you able to verify this perhaps in your environment? That would be very helpful if possible.

[AstroDogeDX]

@Pringled After running:
pip install 'semble'
pip install 'semble[mcp]'
...then trying:
uvx semble
...I'm met with:

resource module not available on Windows
MCP dependencies are not installed. Run: pip install 'semble[mcp]'

pip throws no errors during the installation after both commands, including the MCP dependencies, but yes I also get the 'not installed' message. Happy to continue poking if you tell me where to look.

---

Running on Windows 10 Pro (19041.vb_release.191206-1406)
PowerShell 7.5.5 via VSCodium

[Pringled]

@AstroDogeDX I think this issue is that you're installing with pip and then running with uvx. Can you try running uvx 'semble[mcp]' and let me know if that works? I still think the resource thing is that warning from bm25. Alternatively you can install with uv: uv tool install 'semble[mcp]' instead of pip. I will make some updates to our docs to make this distinction a bit clearer as well.