feat(mcp): refactor to use go-sdk

[manusa]

Fixes #219

[matzew]

some of the tests still have reference to this. Will that clean-up follow in a separate PR?

[manusa]

The tests are still running using the Mark3Labs MCP client.
We can eventually move to just use the official go-sdk mcp client, but I find it more reliable this way though.

[matzew]

this is also to follow up, w/ upstream?

[manusa]

yup, I think we discussed this earlier today or yesterday in a different PR or chat.
Once this is in I'll open an issue upstream and see if they are receptive to a SetTools implementation (comments in their code don't suggest they might want it, so better to provide a real use-case/scenario).

[matzew]

wondering: was the old impl than not "compliant" to latest spec?

[manusa]

I think mark3labs was more lenient in this regard.
I'm not sure that the MCP protocol states that listening for tool changes is something that requires a session id, but in the go-sdk implementation it does.

[matzew]

I want to test with w/ the checked in keycloak scripts

[matzew]

I1113 08:35:06.331469   44005 middleware.go:38] "mcp tool call: namespaces_list(map[])"
I1113 08:35:06.331516   44005 manager.go:244] "Authorization header found (Bearer), using provided bearer token"
I1113 08:35:06.342283   44005 middleware.go:29] "POST /mcp 200 20.15874ms"

tested the keycloak setup from the docs folder and it worked!

[matzew]

LGTM 🎉

[manusa]

Merging 🚀