Sentry MCP Server (Official)

What is Sentry MCP Server (Official)?

Sentry MCP Server (Official) is Sentry's own Sentry MCP server that connects AI coding assistants such as Claude Code, Cursor and VS Code to your Sentry errors, issues, traces and performance data through the Model Context Protocol.

It is maintained by Sentry in the getsentry/sentry-mcp repository and runs mainly as a hosted remote service at https://mcp.sentry.dev/mcp, with OAuth sign-in. Sentry says the tool selection is focused on human-in-the-loop debugging workflows, not on exposing every Sentry API. A stdio transport, published on npm as @sentry/mcp-server, is also available and is the easiest way to use it with self-hosted Sentry.

How to use Sentry MCP Server (Official)?

You need a Sentry account. For the hosted server nothing is installed: the first connection opens an OAuth flow in your browser to link your Sentry account.

The base endpoint is https://mcp.sentry.dev/mcp. You can scope it to one organization (https://mcp.sentry.dev/mcp/{organizationSlug}) or one project (https://mcp.sentry.dev/mcp/{organizationSlug}/{projectSlug}). When scoped, tools default to that org or project and unneeded discovery tools are hidden; Sentry recommends project scoping when possible. Add ?experimental=1 to enable experimental tool variants.

Claude Code

Add the remote server over HTTP:

claude mcp add --transport http sentry https://mcp.sentry.dev/mcp/{organizationSlug}/{projectSlug}

Or install the Claude Code plugin, which adds a sentry-mcp subagent that Claude delegates to when you ask about Sentry errors, issues, traces or performance:

claude plugin marketplace add getsentry/sentry-mcp
claude plugin install sentry-mcp@sentry-mcp

Cursor

Use the "Install MCP Server" button on mcp.sentry.dev, or add this to your MCP settings:

{
  "mcpServers": {
    "sentry": {
      "url": "https://mcp.sentry.dev/mcp/{organizationSlug}/{projectSlug}"
    }
  }
}

VS Code

Open the Command Palette, run "MCP: Add Server", choose HTTP and enter https://mcp.sentry.dev/mcp/{organizationSlug}/{projectSlug}.

Claude Desktop and other stdio clients

For clients that launch local servers (or for self-hosted Sentry), create a Sentry User Auth Token with the scopes org:read, project:read, project:write, team:read, team:write and event:write, then use the stdio package:

{
  "mcpServers": {
    "sentry": {
      "command": "npx",
      "args": ["@sentry/mcp-server"],
      "env": {
        "SENTRY_ACCESS_TOKEN": "your-token",
        "EMBEDDED_AGENT_PROVIDER": "openai",
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

The LLM provider variables are only needed for the AI-powered search tools. You can also run it directly:

npx @sentry/mcp-server@latest --access-token=sentry-user-token

For self-hosted Sentry add --host=sentry.example.com (hostname only), and --insecure-http if the deployment only serves plain HTTP.

Remote with an explicit token

Clients that support custom headers can skip OAuth and pass a Sentry API token:

{
  "mcpServers": {
    "sentry": {
      "url": "https://mcp.sentry.dev/mcp",
      "headers": {
        "Authorization": "Sentry-Bearer ${SENTRY_ACCESS_TOKEN}"
      }
    }
  }
}

Key features of Sentry MCP Server (Official)

• Official server built and maintained by Sentry (getsentry).
• Hosted remote endpoint with OAuth, so there is nothing to install or keep updated.
• Organization- and project-scoped URLs that pre-fill slugs and hide unneeded discovery tools.
• Natural-language search over issues and events (errors, logs, spans, metrics, profiles, replays).
• Seer integration for root cause analysis and suggested code fixes.
• Issue triage actions: resolve, reopen, ignore or assign issues.
• A tool catalog (search_sentry_tools / execute_sentry_tool) for long-tail operations such as releases, teams, DSNs and docs lookup.
• Stdio transport for self-hosted Sentry, with skills that can be enabled or disabled (for example --skills=inspect,seer or --disable-skills=project-management).

Sentry MCP Server (Official) tools

The mcp.so listing shows nine top-level tools:

• Discovery: find_organizations and find_projects look up the org and project slugs other tools need. They are hidden when the URL is already scoped.
• Search: search_issues returns lists of grouped issues (for example is:unresolved level:error), while search_events handles counts, aggregations and individual events across errors, logs, spans, metrics, profiles and replays. Both accept natural language or Sentry search syntax.
• Inspect: get_sentry_resource fetches an issue, event, trace, span, replay, monitor or AI conversation from a Sentry URL or ID.
• Fix and triage: analyze_issue_with_seer runs Seer root cause analysis (a new analysis can take about 2–5 minutes; results are cached), and update_issue changes status or assignment.
• Catalog: search_sentry_tools finds operations that are not exposed as top-level tools, and execute_sentry_tool runs them, for example whoami.

Use cases of Sentry MCP Server (Official)

• Fix a production bug from your editor: "Look at https://my-org.sentry.io/issues/PROJECT-123/ and fix the root cause in this repo."
• Triage new errors: "Show unresolved errors first seen in the last 24 hours in my-project."
• Ask for statistics: "How many errors did we have today, grouped by issue?"
• Get an AI fix proposal: "Run Seer on this issue and suggest a fix."
• Clean up the issue queue: "Resolve PROJECT-123 and assign PROJECT-456 to me."
• Debug performance: "Find the slowest database spans in the last hour."

Sentry MCP Server (Official) vs alternatives

• Hosted remote vs stdio: the remote server at mcp.sentry.dev uses OAuth and needs no LLM key for search. The stdio package needs a user auth token, and an LLM provider key for the AI-powered search tools, but it can target self-hosted Sentry.
• Claude Code plugin vs plain MCP: the plugin wraps the same server in a dedicated subagent; claude mcp add exposes the tools directly to the main agent.
• Community Sentry servers: some directories also list smaller community Sentry MCP projects. The official getsentry server is the one Sentry maintains and documents.

Troubleshooting Sentry MCP Server (Official)

• AI search tools are missing (stdio): natural-language search needs an LLM provider (OpenAI, Azure OpenAI, Anthropic or OpenRouter). Set its API key and set EMBEDDED_AGENT_PROVIDER explicitly; other tools work without it.
• Seer tools missing on self-hosted: Seer is not part of self-hosted Sentry, so the seer skill is dropped when --host points at a non-sentry.io host. If your deployment runs Seer, add --skills=inspect,seer.
• Plain-HTTP self-hosted instance won't connect: add --insecure-http, and pass only the hostname to --host.
• Permission errors with a token: make sure the user auth token has the scopes listed above.
• OAuth problems in MCP Inspector on 127.0.0.1: use localhost instead.