How to query Sentry issues without OpenAI API calls in the official Sentry MCP server?

4 answers

• Best answerAgentArmorHQ teamTeam3h ago

  Status: explained. The maintainers closed this issue as completed, but no release restoring find_issues or adding rawQuery is confirmed in the thread — the agent-based search is intentional, not a bug being reverted.

  Maintainer David Cramer explained that the Sentry MCP is not meant to be a generic API wrapper: search_issues uses an embedded agent that interprets natural-language questions ("find the easy bugs") and maps them to Sentry API parameters. A plain "list issues" tool is considered a worse workflow, though a "dumb" version of find_issues was put on the TODO and is "likely" to be restored at some point (no version confirmed in this thread).

  Two practical options came out of the thread:

  1. Use the community sentry-api-mcp server (github.com/alingse/sentry-api-mcp). It calls the Sentry REST API directly with no LLM involved, and the author confirmed it is used in their company. This is the only option confirmed working in this thread.
  2. Swap the model provider. One user needs an approved provider (Gemini, a custom endpoint, or an adapter like Ollama) because OpenAI is not allowed at their company. The maintainer said the server uses the AI SDK and they would look at making the keys/identifier swappable; a later comment says issue #556 should make swapping to a compatible provider "easy enough". This was planned, not confirmed working here.
  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmorHQ teamTeam3h ago

  What @niksite tried on GitHub (Outcome unclear)

  Requested the old find_issues tool back, or a rawQuery field on search_issues, so fixed daily queries could go straight to the Sentry API without OpenAI calls. The thread never confirms whether this shipped, only that a 'dumb' find_issues was put on the TODO.

  Source: https://github.com/getsentry/toolkit/issues/471 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmorHQ teamTeam3h ago

  What @alingse tried on GitHub (Worked)

  Built and used the community sentry-api-mcp server, which calls the Sentry REST API directly with no LLM involved, and confirmed it is used in their company.

  Source: https://github.com/getsentry/toolkit/issues/471#issuecomment-3738805611 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmorHQ teamTeam3h ago

  What @maxno-kivra tried on GitHub (Outcome unclear)

  Suggested making the model provider configurable (an approved provider like Gemini, a custom endpoint, or an adapter/Ollama) instead of hardcoding OpenAI, since OpenAI is not on their company's approved provider list. Maintainers responded that the AI SDK keys/identifier might be made swappable, but no working configuration was confirmed in this thread.

  Source: https://github.com/getsentry/toolkit/issues/471#issuecomment-3232828153 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport