Sentry MCP

Official Sentry MCP server: give Claude Code, Cursor, and other AI agents real access to your errors, traces, and triage.

Freemium· Free: $0 USD · Team: $4 USD per user/month · Enterprise: $21 USD per user/monthMCP ServersBring-your-own LLM (OpenAI, Anthropic, Azure OpenAI, or OpenRouter) for natural-language search skills; agent-side model is whatever your MCP client runs.

Visit website →

Best for

Engineering teams already on Sentry who want their coding agent (Claude Code, Cursor, etc.) to debug live incidents against real error and trace data instead of pasted snippets.

Skip if

Teams that do not use Sentry, or that want a fully autonomous agent to close tickets without human review of write actions like resolve/assign.

Sentry MCP is the official Model Context Protocol server from Sentry, giving AI coding agents like Claude Code, Cursor, and any other MCP-capable client structured access to your Sentry organization. Instead of pasting stack traces into a chat window, an agent can call MCP tools to list projects, query issues, fetch full event and trace data, search events with natural language, run triage actions (assign, resolve, comment), and dig through performance and error telemetry alongside the code it is editing.

The server exposes two deployment modes. The hosted remote server at mcp.sentry.dev speaks HTTP with a Bearer token and works out of the box with any MCP client that supports remote servers with headers; it also supports full OAuth. For local or air-gapped use, `npx @sentry/mcp-server@latest --access-token=...` runs a stdio server against sentry.io or a self-hosted Sentry instance, and Claude Code users can install it in one command from the Sentry plugin marketplace. Individual skills can be toggled on or off so you only grant the agent the tools you actually want it to hold.

It is aimed squarely at human-in-the-loop debugging: an engineer asks their agent "why did checkout error-rate spike after the last deploy", the agent pulls the top issues, walks the trace to the offending span, opens the file at the right line, proposes a patch, and files a triage note back in Sentry. Because it plugs into the same MCP ecosystem as filesystem, git, and browser tools, Sentry data becomes just another primitive the agent can compose against.

Editor's take

This is the reference implementation for what a first-party MCP server should look like: hosted and stdio modes, OAuth, self-hosted support, and a skill toggle so you can grant only what you need. If Sentry is already your source of truth for errors, wiring this into Claude Code or Cursor is the highest-leverage MCP install we tested this quarter.

— The AI Tool Bible editorial team

Pros

✅ Official first-party server from Sentry — kept in step with API changes, not a community wrapper
✅ Both a hosted remote endpoint (mcp.sentry.dev, OAuth or Bearer) and a local stdio binary via npx
✅ Works with self-hosted Sentry, not only sentry.io
✅ One-line install for Claude Code through the Sentry plugin marketplace; also documented for Cursor and MCP Inspector
✅ Fine-grained skill toggles let you scope which tools the agent can call
✅ Covers the full debugging loop: issues, events, traces, search, and triage actions like assign/resolve/comment
✅ Open source, so the tool surface and prompts can be audited or extended

Cons

⚠️ You still need a paid Sentry plan to get useful volumes of events, retention, and org seats
⚠️ Natural-language search tools require you to bring your own OpenAI/Anthropic/Azure/OpenRouter key — extra cost and setup
⚠️ Value collapses if your team is not already invested in Sentry as its error/APM backend
⚠️ Remote server is single-tenant per token — sharing across a team means each engineer wires their own auth
⚠️ MCP itself is still young; some clients handle remote servers with headers imperfectly and stdio is often the fallback
⚠️ Read/write tools mean a misbehaving agent can resolve or reassign real issues — human-in-the-loop is not optional

Use cases

Debug production error from IDE agentRoot-cause a latency regression via trace lookupTriage new issues (assign, resolve, comment) from chatCorrelate a failing test with recent Sentry eventsNatural-language search across eventsPost-deploy error-rate investigationPull stack trace and open the offending file for a fixSummarize top issues for a standupQuery self-hosted Sentry from an internal coding agent