Sentry MCP: Debug Production Errors with Claude

Sentry MCP is Sentry’s official Model Context Protocol server, now developed in the getsentry/toolkit repository alongside the Sentry CLI. It is a remote server built for human-in-the-loop coding agents, so its tools focus on debugging: issues, events, stack traces, traces, logs and performance data, rather than every corner of the Sentry API.

Who it is for

Developers who use Sentry for error monitoring and want Claude Code to look at the real production error before proposing a fix.

Install

The hosted service lives at mcp.sentry.dev. The repo documents a Claude Code plugin that adds a sentry-mcp subagent Claude delegates to when you ask about errors:

claude plugin marketplace add getsentry/sentry-mcp
claude plugin install sentry-mcp@sentry-mcp

For self-hosted Sentry, run the stdio transport with a user auth token:

npx @sentry/mcp-server@latest --access-token=sentry-user-token

Add --host=sentry.example.com for a self-hosted instance. Remote clients that support headers can pass a token with Authorization: Sentry-Bearer ....

Use cases

• Fix with evidence: “Look at the latest unresolved issue in the checkout project and find the root cause.”
• Trace analysis: ask why an endpoint got slow, using trace data.
• Release checks: list new issues introduced since the last deploy.
• Triage: summarize the top errors this week by user impact.

Pros

• Official and focused on real debugging workflows.
• Remote OAuth for Sentry SaaS, stdio for self-hosted.
• Plugin route gives you a dedicated subagent, keeping Sentry output out of your main context.

Cons and caveats

• AI search tools need a configured LLM provider (OpenAI, Azure OpenAI, Anthropic or OpenRouter).
• FSL license, which restricts competing commercial use.
• Stack traces and event payloads can be large; ask for summaries first.

Links

• Repo: getsentry/toolkit
• Service: mcp.sentry.dev
• Related: Sentry skills, Chrome DevTools MCP, subagents guide