@sentry/mcp-server

What it does

The server bridges Claude, Cursor, and similar coding agents to Sentry's API, focusing on debugging workflows. It exposes tools for querying issues, events, and traces from a Sentry instance (SaaS or self-hosted), and includes natural-language search capabilities powered by an embedded LLM agent (OpenAI or Anthropic). The server operates both as a remote MCP endpoint and via stdio transport, adapting to different deployment models.

Who it's for

Engineering teams integrating Sentry error tracking directly into AI-assisted development workflows. Particularly useful for developers using Claude Code or Cursor who want to debug production issues without leaving their IDE—asking the agent about specific errors, stack traces, or performance regressions.

Common use cases

• Query Sentry issues and error details directly from Claude Code while debugging
• Search for past occurrences of a newly reported error to understand its scope
• Retrieve performance traces and transaction data for latency investigation
• Gather issue context (comments, attachments) for in-context reasoning by the agent
• Automatically fetch error resolution history to inform fix strategies

Setup pitfalls

• 14 secrets discovered in code/config: Carefully audit credentials before deploying. Never commit SENTRY_ACCESS_TOKEN, OPENAI_API_KEY, or ANTHROPIC_API_KEY to version control; use environment variables or secret management.
• AI search tools require explicit LLM provider: Set EMBEDDED_AGENT_PROVIDER to openai or anthropic and provide the matching API key; without it, search tools are unavailable, though read operations still work.
• Self-hosted Sentry requires extra flags: Use --host and --insecure-http for internal HTTP-only deployments, and --disable-skills=seer if your instance doesn't support Sentry's ML features.
• No CI pipeline: Test against your Sentry instance before production use, as the project lacks automated CI coverage.