getsentry/sentry-mcp — reverse-engineered prompt
Reverse engineered prompt
Build me a TypeScript MCP service that lets coding agents connect to Sentry and help developers debug errors, issues, traces, and performance problems from inside tools like Cursor or Claude Code.
I want it to work as a remote MCP server with Sentry OAuth for normal hosted use, and also have a simple stdio mode where someone can pass a Sentry auth token for local or self hosted Sentry installs. It should expose practical tools for developer workflows, like finding projects, reading issues and events, searching Sentry data in plain English when an OpenAI or Anthropic key is configured, and safely sending updates back to Sentry when allowed.
Please include clear setup docs, example MCP config, environment variables, a local dev server, and a way to test it with the MCP inspector and a small CLI. Keep the experience focused on helping an AI coding assistant understand what went wrong in Sentry and guide the developer toward a fix. Look up current MCP and Sentry docs online if you need to.
Want more depth? Deep Reverse