Evaluate using tmattio/ocaml-mcp

[mseri]

Currently using https://tangled.sh/@anil.recoil.org/ocaml-mcp
Evaluate pros and cons of using https://github.com/tmattio/ocaml-mcp

[lukstafi]

Neither of them is available in the opam repository...

[mseri]

I vendor them for now :P

[mseri]

A partly working implementation using the second is here: https://github.com/mseri/snf-mcp/tree/tmattio

It does not work with all clients though, while the current main branch is working fine for me on all clients I have tried

[mseri]

@tmattio and @avsm what would you suggest for an alpha release on the opam repository? Should we change the library names to avoid clashes and release both? Only one?

---

I don't have time to investigate why @tmattio library, which is more comprehensive, is not fully working for me until after the end of January.

What I had inferred last time I tried, was that in stdio mode my clients were not happy to call the initialize method and chain it with the rest of the calls, and were instead expecting the call to work without it. My tests were with Jan, LMStudio and a plugin of the python llm client (that is, the clients I normally use myself). The current main branch so-far has been working perfectly fine with all three.

The other thing I wanted to investigate before switching branches, is if we could avoid the two Obj.magic calls in https://github.com/tmattio/ocaml-mcp/blob/49ee348859d8167efb23dfad714250e92177d564/lib/mcp-sdk/common.ml#L322 and https://github.com/tmattio/ocaml-mcp/blob/49ee348859d8167efb23dfad714250e92177d564/lib/mcp-sdk/common.ml#L520