Sentry MCP Server

Official

Errors with full context — stack traces, issue triage and AI-powered root-cause analysis from Sentry's server.

Local + remotestdiostreamable-httpTypeScriptMIT 400

What is the Sentry MCP server?

The Sentry MCP server ends the stack-trace-paste ritual. Instead of copying error text into a chat window (losing breadcrumbs, tags, release context and frequency data in the process), your assistant queries Sentry directly: projects, issues, events with complete context — and can summon Seer, Sentry's own root-cause analysis agent, for the hard ones.

Debugging-by-URL is the signature move. Paste a Sentry issue link and ask "what's going on here?" — the assistant pulls the full event: exception chain, breadcrumbs leading in, affected release, user impact counts. With repo access alongside, the loop closes: error understood, offending code found, fix drafted, all in one thread. The difference between this and paste-the-traceback is the difference between reading a police report and visiting the scene.

Triage scales the same way: "what's new in production since yesterday's release?" or "which issues affect the most users this week?" become direct queries rather than dashboard sessions. For on-call, asking about a spike beats clicking through it.

The hosted server (mcp.sentry.dev, OAuth) is the recommended path and covers SaaS and self-hosted Sentry; a local stdio mode exists for the token-inclined. Access follows your Sentry permissions. The tool count is substantial (projects, issues, events, releases, DSNs, Seer invocation), but the daily gesture is simple: give it an error, get back understanding — which is, after all, what Sentry was always for.

Triage habits worth building

  • Lead with the URL whenever you have one. A Sentry issue link already encodes org, project and issue ID, so it needs no supporting explanation to land on the right event.
  • update_issue resolves and assigns, which folds the bookkeeping half of triage into the same conversation instead of leaving a second pass in the UI for later.
  • Seer runs asynchronously — begin_seer_issue_fix starts it and the status tool collects the analysis. Kick it off at the start of a session and read the findings once you have the basics yourself.
  • Scope the OAuth grant to the organisation you are actually debugging if you belong to several. Cross-org issue searches return noise that is tedious to filter back out.

What you can do with it

Debug from a link

Paste a Sentry issue URL and get the full-context investigation.

Release health checks

"What broke since we shipped?" answered from real error data.

Root-cause on demand

Seer analysis plus your repo-connected assistant closes error-to-fix.

Available tools

ToolWhat it does
find_issues / get_issue_detailsQuery issues and read full event context
find_projects / find_releasesProject and release enumeration
get_event_attachmentFetch event attachments
update_issueResolve, assign and manage issues
begin_seer_issue_fix / get_seer_issue_fix_statusInvoke Sentry's Seer root-cause agent
create_dsn / find_dsnsManage project DSNs

How to install the Sentry MCP server

claude mcp add --transport http sentry https://mcp.sentry.dev/mcp

Configuration

A Sentry account (SaaS or self-hosted). OAuth for the hosted server; a user auth token for local mode.

VariableDescriptionRequired
SENTRY_ACCESS_TOKENUser auth token for local mode (remote uses OAuth)Optional
SENTRY_HOSTSelf-hosted Sentry hostname if applicableOptional

Example prompts to try

  • Investigate this Sentry issue and propose a fix: [issue URL]
  • Which production issues affected the most users this week?
  • Run Seer on the checkout NullPointerException and summarise its analysis.

Frequently asked questions

Yes — the hosted MCP server supports both SaaS and self-hosted backends, and local mode can point at your instance via SENTRY_HOST.