Skip to main content
← Back to Articles
mcpcursorsentrymonitoringdebuggingsetup2026

Sentry MCP Server Cursor IDE Setup 2026: mcp.sentry.dev OAuth Config

Set up Sentry's official remote MCP server in Cursor IDE: the mcp.sentry.dev/mcp endpoint, OAuth sign-in, the search_issues and Seer root-cause tools, and the token-based stdio fallback for CI or locked-down machines.

By Web MCP GuideAugust 26, 20269 min read

How do you set up the Sentry MCP server in Cursor? Add a sentry entry under mcpServers in ~/.cursor/mcp.json with a url field pointing at https://mcp.sentry.dev/mcp, restart Cursor, and complete the OAuth sign-in that opens in your browser on first connect. No API token to generate or paste for this path — authentication happens through the same login you already use for the Sentry dashboard.

This guide covers Sentry's official hosted MCP server. There's also a local stdio variant (sentry-mcp-stdio / mcp-server-sentry) authenticated with a token instead of OAuth, covered in Step 4, for CI environments and machines that can't open a browser for the consent flow.

Quick Reference

MaintainerSentry (official)
Remote URLhttps://mcp.sentry.dev/mcp
Auth (remote)OAuth via browser
Auth (stdio fallback)Sentry auth token
Cursor version needed0.47+ for url-based remote MCP entries
Setup timeUnder 5 minutes

Prerequisites


  • Cursor IDE 0.47 or later for the remote url path (older builds need the mcp-remote proxy pattern — see Step 4's note).

  • A Sentry account with access to the organization and projects you want Cursor to query.

  • A browser available on first connect, to complete the OAuth consent screen.
  • Step 1: Add the Remote Server to Cursor

    Open ~/.cursor/mcp.json (global) or .cursor/mcp.json (project-scoped):

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

    That's the entire config — no command, no args, no env block for the OAuth path. Restart Cursor, or reload the window, and check Settings → MCP to confirm the server is listed.

    Step 2: Complete OAuth Sign-In

    The first tool call — or the first time you open Cursor's MCP panel after adding the entry — triggers a browser window prompting you to log in to your Sentry organization and authorize the connection. Approve it. Cursor stores the resulting token and reuses it for subsequent sessions; you shouldn't need to repeat this unless the token is revoked or expires.

    If you belong to more than one Sentry organization, the consent screen is where you pick which one this connection covers. A single MCP entry is scoped to one org at a time — add a second entry with a different key if you genuinely need Cursor talking to two organizations side by side.

    Step 3: Test the Connection

    In Cursor chat:

    Find unresolved Sentry issues affecting more than 10 users in the last 7 days.
    

    Or, once you know a specific issue exists:

    Get the details for Sentry issue PROJECT-123, including the stack trace and the most recent event.
    

    A response with real issue titles, user-impact counts, and stack frames means you're connected. If Cursor says it can't find the tool, or nothing happens, check Output → MCP Logs before re-editing the JSON.

    Step 4: Token-Based stdio Alternative (CI, Locked-Down Machines)

    Use this only when a browser-based OAuth flow genuinely isn't possible — a CI runner, a bot account, or a machine with browser access blocked by policy. Generate a Sentry auth token from your account settings, then configure the local uvx-based server instead of the remote URL:

    {
      "mcpServers": {
        "sentry": {
          "command": "uvx",
          "args": ["mcp-server-sentry", "--auth-token", "YOUR_SENTRY_TOKEN"]
        }
      }
    }
    

    Do not hardcode the token directly in a mcp.json you might commit to a shared repo — reference it from an environment variable instead, or keep this config in your global (not project) file. On older Cursor builds that don't support a bare url entry for the remote server, the same token-based stdio path is also the practical fallback, since it doesn't depend on Cursor's newer remote-MCP support.

    Step 5: Practical Workflows

    Fixing a production error without leaving the editor

    Pull the full stack trace and breadcrumbs for the top unresolved issue in the payments-api project, then look at the relevant source file and tell me what's actually causing it.
    

    Triage sweep at the start of a shift

    List every unresolved issue across my Sentry projects that started in the last 24 hours, sorted by number of affected users, and summarize the top 3.
    

    Using Seer for root-cause analysis

    Trigger Seer analysis on issue PROJECT-456 and summarize what it identifies as the likely root cause before I start debugging manually.
    

    Correlating a release with a spike

    Did the error rate for the checkout-service project increase after the release that shipped yesterday? Pull the relevant issues and tell me if they trace back to that deploy.
    

    Troubleshooting

    OAuth popup never appears
    Confirm your Cursor version supports url-based remote MCP servers (0.47+). If it does and nothing happens, check that your network isn't blocking mcp.sentry.dev, then remove and re-add the server entry to force a fresh connection attempt rather than reusing a stalled one.

    Tools don't appear in Agent mode after adding the config
    MCP config changes require a full Cursor restart, not just a window reload — this is the single most common reason a freshly added server doesn't show up. Fully quit and relaunch.

    "Unauthorized" after OAuth completed successfully
    The stored session may have been revoked from Sentry's side (an org admin rotated app access, or the token expired). Remove the server entry and re-add it to trigger a fresh OAuth flow rather than trying to debug a stale token.

    Issues come back but stack traces are missing frames
    This usually means the project's source maps or debug symbols weren't uploaded for that release — a Sentry-side configuration issue, not an MCP connection problem. Check the issue in the Sentry dashboard directly; if frames are missing there too, the MCP server is faithfully reporting what Sentry has.

    Seer analysis tool isn't available
    Seer is a Sentry feature with its own plan and organization-level enablement requirements. If begin_seer_issue_fix doesn't appear in your tool list, confirm Seer is turned on for your organization in Sentry's dashboard — it's not something the MCP config controls.

    Token-based stdio server works locally but fails in CI
    Confirm the environment variable holding the token is actually exported in the CI runner's environment and not just present in a local .env file that CI doesn't load. Also confirm uvx (or Python/uv) is available on the CI image — it isn't preinstalled on most minimal runners.

    When Not to Use This

    Be careful about auto-approving write-capable Sentry tools (resolving issues, assigning them, muting alerts) in an unattended agent loop — an agent that marks a real production issue as resolved because it "looks fixed" from a code read is a bad failure mode. Keep manual confirmation on for anything beyond read-only issue lookups.

    Frequently Asked Questions

    Q: Do I need a Sentry auth token for the recommended setup?
    A: No. The recommended path is the remote OAuth server at mcp.sentry.dev/mcp — you sign in through a browser the same way you'd log into the Sentry dashboard. Auth tokens are only needed for the stdio fallback covered in Step 4, used for CI or machines that can't complete a browser OAuth flow.

    Q: Can Cursor see issues across all my Sentry projects, or do I need to configure each one?
    A: All projects your authenticated account can access are queryable — there's no per-project config entry to add. Scope individual requests with a project name in your prompt ("in the payments-api project") when you don't want a cross-project search.

    Q: What's the difference between this and just pasting an error message into chat?
    A: Pasting an error gives Cursor exactly what you copied and nothing else. The MCP connection lets it pull the full stack trace, breadcrumbs, affected-user counts, and related events directly from Sentry, and cross-reference that against your actual source files — a materially richer starting point for debugging.

    Q: Is the Seer tool part of the base MCP server, or a separate product?
    A: Seer is a Sentry AI feature with its own enablement and plan requirements. The begin_seer_issue_fix tool is exposed by the MCP server when Seer is available on your organization, but it's a Sentry product capability, not something this integration adds independently.

    Q: Does this work with self-hosted Sentry, or only Sentry's cloud product?
    A: This guide covers Sentry's cloud-hosted mcp.sentry.dev endpoint. Self-hosted Sentry instances would need their own MCP server deployment if Sentry documents one for that configuration — check your self-hosted version's own docs before assuming the cloud endpoint applies.

    Related Guides


  • Datadog MCP Server: Cursor IDE Setup (2026)

  • Honeycomb MCP Server: Cursor IDE Setup (2026)

  • New Relic MCP Server: Cursor IDE Setup (2026)

  • Debugging MCP Server Issues in Cursor

  • MCP Security Best Practices
  • Official docs cited


  • Fix Production Issues in Cursor with Sentry MCP + Seer

  • Sentry MCP and Cursor Setup

  • sentry-mcp-stdio (official repo)

  • Cursor MCP reference

  • Related guides