How do I get breadcrumbs in Sentry issue details with sentry-mcp?
Using the Sentry MCP server (getsentry/toolkit) with Claude Code, the issue details returned to the agent don't include breadcrumbs, so the agent can't see what user actions led up to a crash. There's no error; the data is simply absent from the response. Is there a way to enable breadcrumbs or fetch them separately?
Originally asked on GitHub: https://github.com/getsentry/toolkit/issues/534
3 answers
Status: the issue was closed as completed, but the thread doesn't name a stable release that ships breadcrumbs in the default issue response.
Maintainers held off adding breadcrumbs by default because the payload can be large. The maintainer's stated plan was to add either a "get lots of breadcrumbs"-style flag or a lightweight, descriptively named tool, and to include a few breadcrumbs (about the last 5) in the default issue response with a hint to a follow-up tool call. Nothing in the thread confirms that plan shipped. On further inspection, the maintainer called recent breadcrumbs "absolutely useless" and was weighing a
get_breadcrumbs_for_issuetool against exposing the Seer toolchain as a remote agent; multiple users +1'd the tool, but no final decision is recorded.The only concrete, currently available step in the thread: a maintainer asked testers to append
?experimental=1to the MCP connection string and report back, especially any prompts that run in circles or fail on the first try. The comment does not say what that flag changes, and no follow-up result appears in the thread, so treat it as experimental and unverified.Workaround that worked for one developer: have a separate agent (Codex) fetch breadcrumbs for a specific issue via the Sentry API instead of through the MCP tool. General advice: if you use the API directly, use a token scoped to the projects you need and avoid putting it in logs.
Did this work for you?What @evelant tried on GitHub (Worked)
Had Codex fetch breadcrumbs for particular issues using the Sentry API directly instead of through the MCP server.
Source: https://github.com/getsentry/toolkit/issues/534#issuecomment-3813513723 (opens external site)
Did this work for you?What @dcramer tried on GitHub (Outcome unclear)
A maintainer asked users to append
?experimental=1to the MCP connection string and report how it goes, especially any prompts that run in circles or fail on the first try. The comment doesn't state what the flag enables, and no confirmation follows in the thread.Source: https://github.com/getsentry/toolkit/issues/534#issuecomment-3874393957 (opens external site)
Did this work for you?
Related questions
- Does getsentry/toolkit support granular token scopes (read-only default + 401 upgrade)?
- How to query Sentry issues without OpenAI API calls in the official Sentry MCP server?
- Sentry MCP search_issues/search_events fail: 'Incorrect API key provided' on mcp.sentry.dev
- Sentry MCP search_issues fails with "OPENAI_API_KEY required" when EMBEDDED_AGENT_PROVIDER is anthropic
- Sentry MCP `update_issue` ignores forever despite "until it escalates" reply