Sentry MCP

When an agent is fixing a bug, Sentry MCP lets it pull the production evidence itself: the issue, the stack trace of a specific event, the trace and the release. Paste a Sentry issue URL into Claude Code or Cursor and the agent can read it before touching code. Sentry scoped it to debugging workflows for human-in-the-loop coding agents.

Setup

• Hosted: claude mcp add --transport http sentry https://mcp.sentry.dev/mcp. The first connection starts an OAuth flow; /mcp/{organizationSlug}/{projectSlug} scopes it to one project.
• Self-hosted Sentry: npx @sentry/mcp-server@latest --access-token=TOKEN --host=sentry.example.com over stdio, with a user auth token that has org:read, project:read, project:write, team:read, team:write and event:write. The AI search tools also need an LLM key and EMBEDDED_AGENT_PROVIDER.

What it exposes

Tools are grouped into skills: inspect, seer, docs, triage and project-management.

• get_issue_details, get_event_stacktrace, get_trace_details, get_sentry_resource
• search_issues and search_events, which turn plain language into Sentry query syntax
• analyze_issue_with_seer for root cause analysis with suggested fixes
• update_issue, add_issue_note, plus project and alert tools

Limits

• Sentry describes the stdio transport as a work in progress.
• Seer isn't part of self-hosted Sentry, so the seer skill is dropped when --host points at your own server.

A fit for teams that already send errors to Sentry and debug with an AI coding assistant.