Sentry MCP

What getsentry/sentry-mcp MCP server does

The getsentry/sentry-mcp MCP server exposes Sentry-focused tools for coding assistants. Its intended use is human-guided debugging rather than broad coverage of every Sentry feature. Agents can use it to investigate application errors, Sentry issues, traces, and performance information during development and incident analysis.

The service is available as a hosted endpoint at https://mcp.sentry.dev/mcp. The repository also provides a stdio transport for local use, including connections to self-hosted Sentry installations. Claude Code can install the repository as a plugin that supplies a Sentry-specific subagent, while Cursor and similar development tools can connect through MCP.

How it works

The remote deployment acts as middleware between an MCP client and the upstream Sentry API. A client can authenticate through the hosted service or pass an upstream Sentry token in the Sentry-Bearer header. That header is separate from the standard Bearer mechanism used for MCP OAuth access tokens, and the service forwards the supplied Sentry token rather than managing its refresh lifecycle.

The local mode runs through the @sentry/mcp-server npm package. It uses a Sentry user auth token and targets Sentry SaaS by default. Set a host override when connecting to a self-hosted installation. Plain-HTTP internal deployments require the insecure HTTP option.

The getsentry/sentry-mcp MCP server includes AI-assisted search functions such as search_events and search_issues. These functions require a configured provider from OpenAI, Azure OpenAI, Anthropic, or OpenRouter because they convert natural-language searches into Sentry query syntax. Without a provider, those particular functions are unavailable, while other tools continue to work.

Setup and configuration

For stdio use, create a Sentry user auth token with these documented scopes: org:read, project:read, project:write, team:read, team:write, and event:write. Start the server with npx -y @sentry/mcp-server@latest and provide the token through the command option or SENTRY_ACCESS_TOKEN.

Set SENTRY_HOST only for self-hosted Sentry. Use MCP_DISABLE_SKILLS to hide selected skill groups, such as seer, when the target installation does not support them. AI search also needs EMBEDDED_AGENT_PROVIDER and the corresponding provider credential. OpenRouter supports optional model and reasoning-effort settings.

The hosted service can receive an explicit token through a custom HTTP header. Its endpoint can also be narrowed with skills or disable-skills query parameters, including examples such as inspect, triage, and seer.

Limitations and notes

The project describes stdio support as still in progress. Some features, including Seer, may not exist on self-hosted Sentry instances. The getsentry/sentry-mcp MCP server therefore supports disabling unsupported skills rather than assuming every deployment exposes the same tools.

Remote token forwarding leaves token lifetime and refresh responsibility with the client or upstream provider. AI-powered search is conditional on external LLM configuration and is not available when no supported provider is configured. The repository also includes development and testing commands, but those are for contributors rather than end-user installation.