Delv
Official (Vendor)Active· 6d2,400t★ 4.3by Sentry

Sentry

Pull live errors, issues, and replays from Sentry into Claude. Great for debugging sessions where the model can see what actually broke in prod.

View repo Homepage
A
Safety & Trust

Delv Safety Grade: A

Score 84/100 · assessed 2026-04-22

Maintainer85
Permissions85
Supply chain90
Transparency80
Incidents100

Sentry's official MCP server provides read-only access to production error data, issues, and session replays. The maintainer is Sentry itself, a well-established observability vendor with strong reputation in the developer tools space. Distribution via npm with standard package management is solid. The server requires a Sentry auth token which grants read access to potentially sensitive production data including stack traces, user sessions, and error context. Permissions are appropriately scoped to read-only API calls, though the data itself can be highly sensitive depending on what your application logs. No known security incidents. Transparency is good with open source code and documentation, though changelog and issue tracking could be more prominent. The main risk is credential management: a leaked SENTRY_AUTH_TOKEN exposes all error data the token can access.

Lethal Trifecta (prompt-injection exposure)

TRIFECTA RISK
All three axes present. This server can read private data, ingest attacker-controlled content, and send data outbound. A poisoned input (a GitHub issue, an email, a webpage) can exfiltrate secrets via this chain. Only install with auditing; avoid on shared or cloud agents.
Private dataYes
Reads secrets, credentials, private files
Untrusted inputYes
Ingests web pages, PRs, issues, emails
External commsYes
Can send data outbound

Error payloads can include attacker-controlled strings (ingested untrusted input), project data is private, outbound via API.

Green flags

  • Official vendor-maintained server from established observability company
  • Read-only API access, no write or delete operations
  • Standard npm distribution with versioning
  • Open source with clear documentation
  • Scoped to single domain (error monitoring)

Red flags

  • Auth token grants access to potentially sensitive production error data
  • Session replays may contain PII or sensitive user interactions
  • Stack traces could expose internal architecture details
  • No apparent token scoping guidance in quick-start docs

Permissions requested

Outbound networkAccess secretsIdentity read
Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Install

npx -y @sentry/mcp-server
Paste into your host's MCP config:
{
  "sentry": {
    "command": "npx",
    "args": ["-y", "@sentry/mcp-server"],
    "env": { "SENTRY_AUTH_TOKEN": "sntrys_..." }
  }
}
Env vars needed: SENTRY_AUTH_TOKEN

Review

Sentry's official MCP server pipes live production errors straight into Claude. You give it a Sentry auth token, and suddenly your AI can query issues, pull stack traces, watch replays, and group errors by root cause without you copy-pasting between tabs. I've used this during post-deploy firefighting. Instead of manually triaging a dozen Sentry alerts, I asked Claude to surface regressions from the past week and group them by likely cause. It came back with a sensible breakdown: three issues stemming from a botched database migration, two from a new API timeout, one from a frontend bundle size spike. That's the kind of pattern-spotting that would have taken me half an hour of clicking through the Sentry UI. The replay integration is particularly clever. If you've got session replay enabled in Sentry, Claude can reference what the user was doing when the error fired. I've used this to connect a vague "undefined is not a function" to a specific interaction sequence, which made the fix obvious. Quirks: it's read-only, so you can't mark issues as resolved or assign them from Claude. The server assumes you know your Sentry organisation and project slugs, which you might not have memorised if you work across multiple teams. And if your Sentry instance is flooded with noise, Claude will dutifully surface that noise too. Garbage in, garbage out. This is for developers who already use Sentry and want to debug faster during incidents or PR reviews. If you're not shipping to prod yet, or you don't have Sentry set up, there's nothing here for you. If you do, and you spend time triaging errors, this is one of the few MCP servers that genuinely saves time rather than just being a novelty.
Verdict

Install this if you're already using Sentry in production and you want Claude to help triage errors during incidents or retrospectives. Skip it if you're not shipping live code or if you prefer the Sentry web UI for everything. It's a time-saver for teams who debug frequently, not a toy.

Good at

  • Pulls live stack traces and replays directly into Claude, which beats copy-pasting from the Sentry UI during triage.
  • Grouping errors by root cause is genuinely useful for post-deploy retrospectives or spotting patterns in a flood of alerts.
  • Official vendor support means it tracks Sentry's API changes and won't break unexpectedly.
  • Works across Claude Desktop, Cursor, and Windsurf without needing separate builds.

Watch out

  • Read-only: you can't mark issues as resolved, assign them, or update Sentry state from Claude.
  • Assumes you know your Sentry org and project slugs, which aren't always obvious if you work across multiple teams.
  • If your Sentry instance is noisy with low-signal errors, Claude will surface that noise without filtering it for you.
  • No built-in rate limiting guidance, so heavy querying might hit Sentry API limits on free tiers.

Use cases

  • Asking Claude to debug a live error
  • Surfacing a week's worth of regressions
  • Pulling stack traces into a PR review
  • Grouping related errors by root cause

Getting started

1. Generate a Sentry auth token from your account settings with read access to issues and replays. 2. Run `npx -y @sentry/mcp-server` to verify the package installs cleanly. 3. Add the config snippet to your Claude Desktop or Cursor MCP settings, replacing `SENTRY_AUTH_TOKEN` with your actual token. 4. Restart your host, then ask Claude to list recent issues from a specific Sentry project to confirm it's working. 5. Watch out: you'll need to know your organisation slug and project slug. Find these in your Sentry project URL if you're unsure.

Works with

Claude DesktopClaude CodeCursorWindsurf

Similar MCPs