Sentry MCP — Use Cases, Install & Live Demo

Name: Sentry MCP MCP Server
Author: getsentry

Why use it

Key features

Search issues with Sentry's query syntax
Pull full event payload: stack trace, breadcrumbs, request, tags
Release health: regression detection, adoption, crash-free rate
Performance traces with span breakdown
Project + team metadata for cross-project queries

Live Demo

What it looks like in practice

sentry-mcp-getsentry.replay ▶ ready

0/0

Install

Pick your client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "sentry-mcp-getsentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  }
}

Open Claude Desktop → Settings → Developer → Edit Config. Restart after saving.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "sentry-mcp-getsentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  }
}

Cursor uses the same mcpServers schema as Claude Desktop. Project config wins over global.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "sentry-mcp-getsentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  }
}

Click the MCP Servers icon in the Cline sidebar, then "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "sentry-mcp-getsentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  }
}

Same shape as Claude Desktop. Restart Windsurf to pick up changes.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "sentry-mcp-getsentry",
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  ]
}

Continue uses an array of server objects rather than a map.

~/.config/zed/settings.json

{
  "context_servers": {
    "sentry-mcp-getsentry": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "@sentry/mcp-server"
        ]
      }
    }
  }
}

Add to context_servers. Zed hot-reloads on save.

claude mcp add sentry-mcp-getsentry -- npx -y @sentry/mcp-server

One-liner. Verify with claude mcp list. Remove with claude mcp remove.

Use Cases

Real-world ways to use Sentry MCP

Debug a production error end-to-end without leaving Claude

👤 Developers triaging on-call alerts ⏱ ~20 min intermediate

When to use: Sentry alert fires; you need root cause now.

Prerequisites

Sentry auth token — User Settings → Auth Tokens → create with 'project:read', 'event:read', 'org:read'
Token in env — Set SENTRY_AUTH_TOKEN and SENTRY_HOST (sentry.io or self-hosted)

Flow

Find the issue

Use sentry-mcp. Find the most-firing issue in project 'web' from the last 6 hours.✓ Copied

→ Issue ID + count + first/last seen
Pull full context

Get the latest event for that issue — stack trace, breadcrumbs, request data.✓ Copied

→ Full event payload with frames + trail of user actions
Correlate to deploy

What release was running when this first appeared? List the commits in that release.✓ Copied

→ Release tag + commit list
Propose fix

Looking at the stack and the commits — which commit likely caused this? Draft a fix in the relevant file.✓ Copied

→ Specific file + line referenced; concrete patch

Outcome: From alert to candidate fix in minutes, with full chain of evidence.

Pitfalls

Source maps missing — stack frames unhelpful — Verify source map upload in your Sentry releases; otherwise minified frames give Claude nothing

Combine with: github

Pre-flight check before promoting a release

👤 Release engineers, deploy gatekeepers ⏱ ~15 min intermediate

When to use: About to promote a canary release to 100%.

Flow

Compare release health

Use sentry-mcp. Compare crash-free rate of release 1.4.0 vs 1.3.9 over their respective canary windows.✓ Copied

→ Per-release crash-free %, adoption, regression count
Check new errors

List any issues unique to 1.4.0 (not seen in 1.3.9 or earlier).✓ Copied

→ New-issue list with counts
Recommend

Given crash-free is X% and N net-new errors, recommend ship/halt with reason.✓ Copied

→ Clear recommendation with thresholds called out

Outcome: Data-driven promote/halt decision, not gut feel.

Pitfalls

Adoption too low to be statistically meaningful — Claude should flag low-adoption windows as inconclusive — not pretend to know

Investigate a perf regression using transaction traces

👤 Backend engineers chasing latency ⏱ ~25 min advanced

When to use: p95 latency on /checkout jumped after the last deploy.

Flow

Pull a recent slow trace

Use sentry-mcp. Get a trace for transaction 'POST /checkout' with duration > 2s from the last hour.✓ Copied

→ Trace with span breakdown
Find the new span

Compare span list to a similar trace from last week. Which spans are new or longer?✓ Copied

→ Diff: new DB query 'SELECT cart_items', avg 1.4s
Tie to code

What commit added that query? Suggest an index or a query rewrite.✓ Copied

→ Commit + concrete fix proposal

Outcome: Root cause for a perf regression, with code-level fix proposal.

Pitfalls

Trace sampling rate too low — no slow trace caught — Bump sample rate temporarily, or ask Claude to use stats endpoints instead

Combine with: github

Combinations

Pair with other MCPs for X10 leverage

sentry-mcp-getsentry + github

From Sentry stack frame straight to a PR draft

Pull issue X from Sentry, find the offending commit via GitHub, draft a fix as a new branch + PR.✓ Copied

sentry-mcp-getsentry + linear

Auto-file engineering tickets from regressions

For each new release-blocking issue in Sentry, create a Linear ticket with the stack + suggested owner.✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call	Cost
search_issues	project, query, statsPeriod?	Filter issues by Sentry query	1 Sentry API call
get_issue	issue_id	Top-level look at one issue	1 API call
get_event	event_id or 'latest'	Pull a specific occurrence with full context	1 API call
list_releases	project, version?	Cross-release comparison	1 API call
get_trace	trace_id	Performance investigation	1 API call

Cost & Limits

What this costs to run

API quota: Standard Sentry API rate limits — bursts may 429
Tokens per call: Event payload can be 5–20k tokens — pull selectively
Monetary: Free for Sentry developer plan; paid plans have higher rate limits
Tip: Use search_issues to narrow first; only get_event on candidates — not whole result lists

Security

Permissions, secrets, blast radius

Minimum scopes: project:read event:read org:read

Credential storage: SENTRY_AUTH_TOKEN env var (use a scoped, project-bound token)

Data egress: All calls to your Sentry host (sentry.io or self-hosted)

Never grant: project:write or org:write to a chat-driven agent

Event data can contain PII (request bodies, user IDs) — apply Sentry's data-scrubbing rules upstream

Troubleshooting

Common errors and fixes

401 Unauthorized

Token expired or scope wrong — regenerate with the listed read scopes

Verify: curl -H 'Authorization: Bearer $SENTRY_AUTH_TOKEN' $SENTRY_HOST/api/0/projects/

Stack trace shows minified frames only

Verify source map upload in your release pipeline; without maps, Claude can't reason about frames

Verify: Sentry UI → Release → Artifacts

404 on a known issue

Issue is in a project the token doesn't cover — re-scope token to that project

Self-hosted Sentry — connection refused

Set SENTRY_HOST explicitly to your instance URL with protocol

Verify: curl $SENTRY_HOST/api/0/

Alternatives

Sentry MCP vs others

Alternative	When to use it instead	Tradeoff
Datadog MCP	Your team uses Datadog APM/error tracking	Different coverage; APM-first, less issue-focused
Honeycomb MCP	You're an observability-first shop using events not stack traces	No first-class issue grouping; trace-centric

More

Resources

📖 Read the official README on GitHub

🐙 Browse open issues

🔍 Browse all 400+ MCP servers and Skills