Merge with dpella/mcp?

[Tritlo]

Hey!

I was recently allowed to open-source our internal mcp-server library (over at https://github.com/dpella/mcp) and was wondering if we should combine our efforts? I understand adding this library due to the older one lagging behind, but now that it is open source, that should no longer be the case. Of course with both as maintainers on here and hackage

[drshade]

Hey @Tritlo !

Apologies I am traveling for work for a few weeks so a little disconnected right now. I would be very keen to add you as a maintainer if you would be interested!

I havent reviewed your library and would need some time to think about the benefits of merging - on the surface they look quite different in philosophy with mcp-server taking a bit more of a template haskell approach based on an ADT definition - in an attempt to be more declarative and type safe. There is obviously a lower-level set of types and handlers which look probably closer to your library, but i would have to spend some time to remind myself of the details in order to assess fit. Of course there are pros and cons of the approach - but this particular style felt like a good fit for MCP at the time - but of course I am open to better ideas!

In terms of hackage - I would much prefer that our CI process (which was majorly enhanced by @jhrcek recently), also publish to hackage upon a new version bump, as I have some concerns around the process I am using right now - especially if the library gains any popularity!

Apologies again for delayed reply but I will be back to productive again in 2 weeks and will take the time to get my head into these topics above.

All the best
Tom

[Tritlo]

sure! Our library is also very type-based, though we rely more on DerivingVia than template Haskell to speed up compilation, but there's some TemplateHaskell as well. We can easily add TemplateHaskell derivation to MCP as well.

My library also has the benefit of being tested with the Claude Agent SDK, ensuring that we're indeed implementing MCP as understood by claude code and other agents.

As for CI: we build and test on our end, though I am against automatically pushing to hackage. Releases should be manually vetted before release so as to not risk polluting Hackage with broken packages that might inadvertently be picked by a build plan.

[jappeace]

guys, you can combine efforts, but these libraries aren't compatible at all. You'd have to provide some kind of backwards compatibility scheme. (or just abandon one appraoch).

[drshade]

Agree. It isn't clear to me what combining would mean. Why not just leave both independent? They look different enough and probably different use cases apply.

[dpwiz]

At least sharing mcp-types would be nice. Even better - agreeing on something like wai.