Can we have find_issues back? Or, have rawQuery field support.

[niksite]

I have a fixed list of queries that I run every morning. Since these queries are fixed, there is no need for OpenAI API calls on them.

Alternatively, could we add rawQuery field support to the search_issues tool alongside the existing naturalLanguageQuery? This would pass queries directly to the Sentry API without OpenAI API calls.

[dcramer]

Can you give me an example of the workflow? Where/how you're running these?

I'm not opposed to swapping out behavior if an agent isnt available, but we're trying to minimize overload and agent-to-agent is likely to become increasingly present in this mcp server

[niksite]

Every morning I start my day by asking Claude Code to check Sentry for new or repeating issues.

That is a fixed list of requests; there is no problem for me to prepare the exact syntax I want (or just let Claude handle that).

That worked very well until now. Now, for the exact same queries, I would need to create an OpenAI account, pay some per-query fee, and check with my company’s legal department to see if we are even allowed to query the OpenAI API at all.

[dcramer]

Gotcha - so for self hosted. Yeah I'm fine with some cases where we swap tools based on distribution (e.g. a weaker non-agent for this, similar to the old), but likely most newer/advanced tools wont be available without either a remote agent that we build or a local embedded agent. I'll put this on the TODO.

[maxno-kivra]

I second this. In my case we have a pretty strict list of approved AI providers, and OpenAI API isn't part of that. I already have access to Gemini, so I would prefer if I could just use that instead. Another option is to allow specifying the endpoint as well and let me run an adapter/ollama.

[dcramer]

Full support for model configuration here - I will see if there’s an easy way to just make the keys and identifier swappable (since we use the AI SDK)

[dcramer]

FYI #556 should at least let you swap to a compatible provider easily enough. I will likely still restore a "dumb" version of find_issues though.