The Sentry MCP server plugs your error-tracking data into an AI client — issues, stack traces, events, breadcrumbs and performance — so a coding agent can debug production problems with real context. It's an official remote server, so there's nothing to host. Here is the setup.
Setup — official remote server
Sentry hosts the server at https://mcp.sentry.dev/mcp. Add it as a remote/OAuth connector and authenticate once:
{
"mcpServers": {
"sentry": {
"url": "https://mcp.sentry.dev/mcp"
}
}
}
On first use you authorise via Sentry (OAuth or a personal access token). For Claude Code, there's also a plugin that adds a Sentry subagent Claude delegates to automatically:
claude plugin marketplace add getsentry/sentry-mcp
claude plugin install sentry-mcp@sentry-mcp
Restart the client after editing; see MCP config file location.
What it gives your agent
Search issues across projects, fetch event payloads and stack traces, read breadcrumbs, and pull performance data — directly into the model's context while it writes a fix.
Verify
Ask: "What's the most frequent unresolved error in my project this week, and what's the stack trace?" Real issue data back means it's connected.
Common problems
- OAuth fails — make sure you authorise the correct Sentry org; re-run the connector auth.
- No data / empty results — your token/account lacks access to that project.
- Client lacks remote MCP — use a client that supports HTTP/OAuth servers, or the Claude Code plugin.
Going further
Sentry pairs perfectly with the GitHub MCP server — read the error in Sentry, fix it in the repo. Many call it the second server to install after GitHub. Browse the developer tools category or curated loadouts.