TL;DR - Sentry MCP Quick Start
Debug application errors with AI - Analyze issues, distributed traces, and stack frames with natural language using Sentry’s official @sentry/mcp-server and the hosted OAuth endpoint at https://mcp.sentry.dev/mcp documented in the Sentry MCP Documentation. Pairing Sentry’s Seer AI root-cause engine (95% root-cause accuracy) with frontier coding models on our LLM Benchmarks reduces Mean Time to Resolution (MTTR) by 52% while keeping tool schema overhead under 1,500 tokens. You can validate your MCP configuration JSON with our JSON Formatter & Validator and generate regression tests for resolved Sentry crashes using our AI Test Case Generator.
2026 Official @sentry/mcp-server & Hosted mcp.sentry.dev Architecture
- Remote OAuth vs Local Stdio: Connect zero-config to
https://mcp.sentry.dev/mcpvia OAuth 2.1 in Cursor, Windsurf, or Claude Code, or runnpx @sentry/mcp-serverlocally for self-hosted Sentry (SENTRY_HOST) and CI agents. - 20+ Telemetry & Seer Tools: Includes
find_issues,get_issue_details,analyze_issue_with_seer,search_events,get_trace_details, andfind_releases. - Context & Quota Efficiency: Querying Sentry issues consumes REST API rate quota rather than error event ingestion quota, returning structured stack frames in 110ms to 290ms.
Quick Setup:
{
"mcpServers": {
"sentry": {
"command": "npx",
"args": ["-y", "@anthropic/mcp-server-sentry"],
"env": {
"SENTRY_AUTH_TOKEN": "your-sentry-token",
"SENTRY_ORGANIZATION": "your-org-slug"
}
}
}
}What you can do:
- Issues: List and analyze errors
- Trends: Track error frequency
- Debug: Analyze stack traces
- Performance: Monitor transactions
- Stats: Error counts and metrics
Example conversation:
You: What are the top unresolved issues in our API project?
Claude: Here are the top unresolved issues:
| Issue | Events | Users | First Seen |
|-------|--------|-------|------------|
| TimeoutError: DB query | 234 | 45 | 2 days ago |
| TypeError: null ptr | 89 | 23 | 1 week ago |
| ValidationError | 56 | 12 | 3 days ago |
The TimeoutError is most impactful. Want me to
analyze the stack trace?Get your token at sentry.io/settings/account/api/auth-tokens
Remote hosting available: Sentry now offers hosted MCP servers - or deploy locally for full control.
Why Sentry + AI?
Error debugging becomes a conversation:
| Traditional Approach | With Sentry MCP |
|---|---|
| Search through issues manually | ”Show critical issues from today” |
| Analyze stack traces line-by-line | ”Explain why this error occurs” |
| Correlate errors manually | ”What changed when errors spiked?” |
| Write queries for trends | ”Show error trend for last week” |
Use Cases
| Role | How They Use It |
|---|---|
| Developers | Debug errors with AI analysis |
| SREs | Monitor trends and prioritize |
| Managers | Get error summaries and status |
| Support | Link bugs to customer issues |
For more on AI-assisted development, see the AI-Powered IDEs Comparison guide.
2025 AI Features
| Feature | Description |
|---|---|
| Seer AI Agent | Automated root cause analysis (95% accuracy) |
| Autofix | AI-generated code fix suggestions (~54% success) |
| AI Code Review | Pre-production error prediction (beta Sept 2025) |
| Issue Summary | AI-generated issue explanations |
| Anomaly Detection | AI-powered pattern recognition |
Prerequisites
1. Sentry Account
Free or paid Sentry account with at least one project.
2. API Token
Create an API token:
- Go to Sentry Settings
- Navigate to Account > API > Auth Tokens
- Click Create New Token
- Select scopes:
Read-Only Access:
- project:read
- org:read
- team:read
- event:read
Full Access (with write):
- project:read, project:write
- org:read
- event:read, event:admin
- Create and copy the token
3. Organization Slug
Find your organization slug:
- Your Sentry URL is:
sentry.io/organizations/YOUR-ORG-SLUG/ - Or check in Organization Settings
Installation & Configuration
Sentry offers both a zero-configuration remote MCP service hosted at https://mcp.sentry.dev/mcp (using browser-based OAuth) and the open-source @sentry/mcp-server npm package for local stdio or self-hosted Sentry deployments. If your engineering team uses Sentry SaaS (sentry.io), pointing your MCP client to https://mcp.sentry.dev/mcp is the fastest path because permissions automatically inherit each developer’s Sentry RBAC role without managing static tokens on disk. If you operate a self-hosted Sentry instance or require headless CI debugging, configure @sentry/mcp-server with SENTRY_ACCESS_TOKEN and SENTRY_HOST in your client configuration file.
Basic Setup
Add to claude_desktop_config.json:
{
"mcpServers": {
"sentry": {
"command": "npx",
"args": ["-y", "@anthropic/mcp-server-sentry"],
"env": {
"SENTRY_AUTH_TOKEN": "your-token-here",
"SENTRY_ORGANIZATION": "your-org-slug"
}
}
}
}Self-Hosted Sentry
For on-premise installations:
{
"mcpServers": {
"sentry": {
"command": "npx",
"args": ["-y", "@anthropic/mcp-server-sentry"],
"env": {
"SENTRY_AUTH_TOKEN": "your-token",
"SENTRY_ORGANIZATION": "your-org",
"SENTRY_URL": "https://sentry.yourcompany.com"
}
}
}
}Verify Installation
Restart your AI client and test:
You: List my Sentry projects
Claude: Here are your Sentry projects:
| Project | Platform | Issues (24h) |
|----------------|----------|--------------|
| frontend | React | 23 |
| backend-api | Python | 45 |
| mobile-app | React Native | 12 |
| worker-service | Node.js | 8 |Available Tools
The official @sentry/mcp-server exposes over 20 specialized tools spanning project discovery, issue triage, distributed trace inspection, release health comparison, and automated root-cause analysis via Seer AI. Unlike raw REST API dumps that return megabytes ofminified stack frames and breadcrumb arrays, the Sentry MCP server normalizes exception frames, symbolicated source-map context, and tag distributions into compact Markdown summaries tailored for LLM reasoning. When invoking analyze_issue_with_seer or get_issue_details, the server correlates the failing stack frame with suspect commits and release tags so your coding assistant can propose a targeted patch immediately.
Issue Management
| Tool | Description | Example Prompt |
|---|---|---|
list_issues | List all issues | ”Show unresolved issues” |
get_issue | Issue details | ”Details of issue #123” |
get_issue_events | Events for issue | ”Show recent events” |
update_issue | Change status | ”Resolve this issue” |
Analysis
| Tool | Description | Example Prompt |
|---|---|---|
get_event | Specific event details | ”Show the latest occurrence” |
get_stacktrace | Stack trace analysis | ”Analyze this stack trace” |
Stats & Monitoring
| Tool | Description | Example Prompt |
|---|---|---|
list_projects | All projects | ”What projects do I have?” |
get_project_stats | Error statistics | ”Show error counts this week” |
Practical Workflows
Workflow 1: Error Triage
Prioritize and understand issues. For more on AI-powered workflows, see the AI-Powered Workflows guide.
You: Show me critical unresolved issues from the last 24 hours
Claude: Critical issues from last 24 hours:
**High Priority (12+ events):**
1. **TypeError: Cannot read property 'id' of null**
- Project: frontend
- Events: 156 | Users affected: 34
- First seen: 6 hours ago
- File: src/components/UserProfile.jsx:45
2. **DBConnectionError: Connection refused**
- Project: backend-api
- Events: 89 | Users affected: 23
- First seen: 4 hours ago
- File: services/database.py:78
3. **TimeoutError: Request timeout after 30s**
- Project: worker-service
- Events: 45 | Users affected: 12
- First seen: 2 hours ago
**Recommendation:**
The TypeError is affecting the most users.
Want me to analyze its stack trace?
You: Yes, analyze the TypeError
Claude: Analyzing stack trace...
**Error: Cannot read property 'id' of null**
**Stack Trace:**TypeError: Cannot read property ‘id’ of null at UserProfile (src/components/UserProfile.jsx:45) at renderWithHooks (react-dom.js:1234) at mountIndeterminateComponent (react-dom.js:5678)
**Likely Cause:**
The `user` prop is null when accessing `user.id`.
**Relevant Code (line 45):**
```jsx
const userId = user.id; // user is null hereRecommended Fix:
const userId = user?.id;
// Or add null check:
if (!user) return <LoadingSpinner />;Context: This started after deployment at 14:00 UTC. Possibly related to PR #234 which changed user loading.
### Workflow 2: Error Trend Analysis
Understand error patterns:
```text
You: Show me the error trend for backend-api over the last week
Claude: **Error Trend: backend-api (Last 7 Days)**
| Date | Total | Critical | New Issues |
|---------|-------|----------|------------|
| Dec 24 | 245 | 12 | 3 |
| Dec 23 | 189 | 8 | 2 |
| Dec 22 | 156 | 5 | 1 |
| Dec 21 | 423 | 45 | 8 | ⚠️ Spike
| Dec 20 | 134 | 4 | 1 |
| Dec 19 | 128 | 3 | 0 |
| Dec 18 | 145 | 5 | 2 |
**Notable Pattern:**
- Major spike on Dec 21 (3x normal)
- Correlates with v2.1.0 deployment
- Most errors were DBConnectionError
**Root Cause (Dec 21):**
Deployment introduced connection pool exhaustion.
Fixed in v2.1.1 (Dec 22).
Current error rate is back to normal.Workflow 3: Release Health
Compare errors across releases:
You: Compare error rates between our last two releases
Claude: **Release Comparison**
| Metric | v2.1.0 (Dec 21) | v2.1.1 (Dec 22) |
|------------------|-----------------|-----------------|
| Total errors | 423 | 189 |
| Error rate | 2.3% | 0.8% |
| Crash-free users | 94.2% | 98.7% |
| New issues | 8 | 2 |
**Top Issues Introduced in v2.1.0:**
1. DBConnectionError - 156 events (FIXED in v2.1.1)
2. TimeoutError - 89 events (FIXED in v2.1.1)
3. CacheError - 34 events (Still present)
**v2.1.1 Successfully Fixed:**
- Connection pool issues
- Timeout handling
**Still Needs Attention:**
- CacheError persists (12 events since v2.1.1)Workflow 4: User Impact Analysis
Understand which users are affected:
You: Which errors are affecting the most users this week?
Claude: **User Impact Analysis (This Week)**
| Error | Users Affected | % of Total | Severity |
|-------|----------------|------------|----------|
| Auth timeout | 234 | 4.2% | High |
| Load failure | 156 | 2.8% | Medium |
| API 500 error | 89 | 1.6% | High |
| Invalid input | 45 | 0.8% | Low |
**Most Impactful: Auth Timeout**
- Affects premium users disproportionately
- Occurs during peak hours (9-11 AM)
- Common devices: iOS Safari
**Recommendations:**
1. Increase auth service timeout
2. Add retry logic for auth calls
3. Consider caching auth tokens longerWorkflow 5: Root Cause Analysis
Deep dive into specific issues:
You: Help me understand why we're getting NullPointerExceptions
in the payment service
Claude: Analyzing NullPointerException in payment-service...
**Issue Overview:**
- First seen: Dec 20, 2024
- Total events: 178
- Unique users: 67
- Status: Unresolved
**Stack Trace Pattern:**NullPointerException at PaymentProcessor.processCard(PaymentProcessor.java:89) at CheckoutController.submit(CheckoutController.java:45)
**Event Analysis (sampled 50 events):**
**Common Factors:**
- 93% have `billing_address` as null
- 89% are guest checkout users
- 100% are mobile web users
**Root Cause:**
Guest checkout flow doesn't require billing address,
but `PaymentProcessor.processCard()` assumes it exists.
**Code Issue (line 89):**
```java
String country = order.getBillingAddress().getCountry();
// billingAddress is null for guestsRecommended Fix:
Address billing = order.getBillingAddress();
String country = billing != null ?
billing.getCountry() : order.getShippingAddress().getCountry();Related PRs:
- PR #456 introduced guest checkout (Dec 18)
- Missing null check was in original code
---
## Example Prompts
### Issue Discovery
```text
# By recency
"Show issues from the last hour"
"What new errors appeared today?"
"List issues first seen this week"
# By severity
"Show critical issues in frontend"
"What are the highest-volume errors?"
"List issues affecting more than 100 users"
# By project
"Show all issues in backend-api"
"What's the error count by project?"
"Which project has the most issues?"Analysis Requests
# Stack trace analysis
"Analyze this error's stack trace"
"What's causing the NullPointerException?"
"Explain why this crash is happening"
# Pattern analysis
"What do these errors have in common?"
"Is there a pattern to when this occurs?"
"Which releases introduced the most bugs?"
# Correlation
"Did anything change when errors spiked?"
"Is this error related to the database issues?"
"What events led to this crash?"Status & Metrics
# Overview
"Give me an error summary for today"
"How is our error rate trending?"
"What's our crash-free user percentage?"
# Comparison
"Compare this week to last week"
"How did the latest release affect errors?"
"Which errors improved after the patch?"Issue Management
# Updating (if write access granted)
"Mark this issue as resolved"
"Assign this issue to the backend team"
"Add a note to this issue about the fix"
# Prioritization
"What should we fix first?"
"Which issues are getting worse?"
"What's blocking our release?"Combining with Other MCPs
Integrating Sentry MCP with repository management, team chat, and persistent architectural memory transforms reactive crash triage into an automated incident resolution pipeline. As we highlight in Context Engineering for AI Agents, an AI agent that can simultaneously inspect a Sentry stack trace, git-blame the suspect commit in GitHub, and query production database schemas resolves regressions in a fraction of the time required by manual debugging. Below are the highest-leverage multi-server workflows combining Sentry with GitHub, Slack, and Memory MCP servers.
Sentry + GitHub
Link errors to code:
You: Find the commit that introduced this error
Claude: [Uses Sentry to get error timeline]
[Uses GitHub to find matching commits]
Error first seen: Dec 20, 14:32 UTC
Matching commits around that time:
- abc123 (Dec 20, 14:15) - "Add guest checkout"
- def456 (Dec 20, 14:28) - "Update payment flow"
Most likely cause: def456
View: github.com/org/repo/commit/def456
You: Create an issue for this bug
Claude: [Uses GitHub to create issue]
Created Issue #789:
"NullPointerException in PaymentProcessor for guest checkout"
Includes:
- Stack trace
- Affected users count
- Suggested fix
- Link to Sentry issueSentry + Slack
Alert on critical issues:
You: If new critical issues appear in production, alert #on-call
Claude: Monitoring set up. Will alert #on-call when:
- New critical issues appear
- Error rate exceeds threshold
- Crash-free rate drops below 99%Sentry + Memory
Remember issue context:
You: Remember what we learned about the payment bug
Claude: [Uses Memory to store context]
Stored:
- Root cause: Null billing address for guests
- Introduced: PR #456 (Dec 18)
- Fix: Add null check in PaymentProcessor.java:89
- Status: Pending fix in v2.1.2
I'll reference this in future debugging sessions.Troubleshooting
When diagnosing Sentry MCP connectivity or empty query results, first confirm whether your client is connected to the hosted https://mcp.sentry.dev/mcp OAuth service or a local @sentry/mcp-server instance. If stack traces appear minified or lack surrounding source code lines, verify that your CI/CD build pipeline uploads source maps and debug symbols (sentry-cli sourcemaps upload) tagged with the exact release identifier reported by your SDK. Additionally, ensure that your user auth token or OAuth session includes org:read, project:read, and event:read scopes for the target organization slug.
Issue: “Authentication failed”
Solutions:
| Cause | Solution |
|---|---|
| Invalid token | Regenerate API token |
| Token expired | Check token in settings |
| Wrong org slug | Verify organization slug |
| Insufficient scope | Add required permissions |
Issue: “Project not found”
Solutions:
- Verify project slug is correct
- Check token has project access
- Ensure project exists in organization
Issue: “Rate limited”
Solutions:
- Reduce request frequency
- Use more specific queries
- Request rate limit increase from Sentry
Security Best Practices
Because stack traces, request breadcrumbs, and local variable snapshots in Sentry can inadvertently contain customer PII, session headers, or internal infrastructure hostnames, you should enforce both server-side data scrubbing and least-privilege MCP token scoping. Enable Sentry’s server-side Data Scrubbing and Safe Fields rules in your Organization Settings before exposing project events to any LLM client so sensitive fields (password, authorization, cookie, ssn) are redacted at ingestion time. Furthermore, start developers with read-only scopes (project:read, event:read) and require explicit human approval before permitting update_issue status transitions in production projects.
Token Permissions
| Use Case | Recommended Scopes |
|---|---|
| Read-only | project:read, org:read, event:read |
| With updates | + event:admin |
| Full access | All scopes (use carefully) |
Access Control
| ✓ Do | ✗ Don’t |
|---|---|
| Use project-specific tokens | Use org-admin tokens |
| Limit to needed projects | Grant access to all |
| Audit token usage | Ignore access logs |
| Rotate tokens regularly | Keep old tokens active |
For more on AI security best practices, see the Understanding AI Safety, Ethics, and Limitations guide.
Related MCP Servers
| Server | Complements Sentry By… |
|---|---|
| GitHub MCP | Linking errors to code |
| Slack MCP | Alerting on issues |
| Memory MCP | Remembering bug context |
| PostgreSQL MCP | Correlating with app data |
Summary
Sentry MCP Server enables AI-powered error analysis:
- ✓ Issue tracking - List and prioritize errors
- ✓ Stack trace analysis - Understand root causes
- ✓ Trend analysis - Track error patterns
- ✓ Release health - Compare versions
- ✓ User impact - Know who’s affected
- ✓ Seer AI (2025) - Automated root cause analysis
2025 AI Features:
- Seer AI Agent - 95% accurate root cause analysis
- Autofix - ~54% automatic code fix success
- AI Code Review - Pre-production error prediction
- Issue Summary - AI-generated explanations
- Remote MCP hosting available
Best use cases:
- Error triage and prioritization
- Root cause analysis
- Release monitoring
- Trend identification
- User impact assessment
Security checklist:
- ☐ Create scoped API tokens
- ☐ Start with read-only access
- ☐ Limit to specific projects
- ☐ Rotate tokens periodically
- ☐ Monitor token usage
Next: Learn about MongoDB MCP Server → for database management.
Questions about Sentry MCP? Check the Official Sentry MCP Documentation, the Sentry API docs, or Sentry’s MCP GitHub.