Sentry

Sentry is the tool your team uses to find out about production errors: when an app fails, Sentry receives the alert, gathers the context, and builds an "issue" with the stacktrace, which user hit it, on which version, how many times it happened.

Sentry's official MCP server is, in the repo's own words, "primarily designed for human-in-the-loop coding agents" — it works as middleware over the Sentry API, optimized for Claude Code, Cursor and similar editors. The idea: investigate and fix errors without bouncing between your IDE and the Sentry panel.

What you can ask

• "What's the latest error in production and which line is firing it?"
• "Summarize the 5 most frequent issues from the last week and group them by file."
• "Go to issue #4521, read me the stacktrace, and tell me which commit most likely introduced it."

The advanced search tools — search_events and search_issues — translate your natural-language question into Sentry's query syntax. For those the server requires an LLM provider (OpenAI or Anthropic) configured; basic tools like get_issue work without an LLM.

What token you need

A Sentry User Auth Token. Generate it at Settings → Account → Auth Tokens → Create New Token in your Sentry account.

Scopes the official server documents as needed for the full tool set: org:read, project:read, project:write, team:read, team:write, event:write.

If you only want to investigate (no mutations), you can start with org:read + project:read + event:read and the server will narrow what it offers. For issue resolution/triage, add the write scopes. The token travels encrypted in your Keychain, never plaintext on disk.