← 16.116.3 →
900 words Slides

16.2 Sentry for Error Monitoring

Course: Claude Code - Enterprise Development

Section: Advanced MCP Integrations

Video Length: 3-4 minutes

Presenter: Daniel Treasure

Opening Hook

"Production failures happen. When they do, you need fast diagnosis and fixes. Today, we're connecting Claude to Sentry to read runtime errors, analyze stack traces, and generate fix recommendations—turning error logs into actionable intelligence."

Key Talking Points

What to say:

  • "Sentry captures runtime errors in your production systems—stack traces, environment data, user context, reproductions."
  • "Manually triaging hundreds of errors is slow. Claude can read Sentry issues and suggest root causes and fixes."
  • "We'll walk through connecting Sentry MCP, reading error context, and having Claude debug failing code paths."
  • "This is read-only access to error data—Claude doesn't modify Sentry, just analyzes."

What to show on screen:

  • Sentry dashboard with real production errors
  • MCP connection configuration
  • Claude analyzing stack traces in real time
  • Generated fix recommendations with code snippets
  • Sentry issue tagging and prioritization

Demo Plan

[00:00 - 00:45] Sentry Setup & MCP Connection 1. Show Sentry dashboard with active project 2. Navigate to Auth Tokens and explain MCP access scope 3. Run claude mcp add command for Sentry 4. Verify connection in Claude Code

[00:45 - 01:45] Reading Error Context 1. Open Claude Code with Sentry MCP connected 2. Ask: "Show me the top 5 errors from the last 24 hours" 3. Claude lists errors with: title, frequency, stack trace, affected users 4. Ask: "What's causing the NullPointerException in AuthService?" 5. Claude reads full stack trace, environment variables, breadcrumbs

[01:45 - 02:45] Root Cause Analysis & Fix Generation 1. Ask Claude: "Analyze the database connection timeout and suggest a fix" 2. Claude reviews stack trace and environment tags 3. Claude identifies: insufficient connection pool size OR network issues 4. Claude generates specific fix (code snippet + config change) 5. Show Claude's reasoning through each hypothesis

[02:45 - 03:30] Prioritization & Triage 1. Ask: "Triage these 10 errors by impact—which should we fix first?" 2. Claude analyzes: error frequency, affected users, severity tags 3. Claude ranks by business impact (not just occurrence count) 4. Show how Claude tags errors for sprint planning

Code Examples & Commands

# Add Sentry MCP adapter
claude mcp add --transport http sentry https://mcp.sentry.dev/mcp

Environment setup:

# Set Sentry auth token (from Sentry dashboard → Auth Tokens)
export SENTRY_AUTH_TOKEN="sntrys_eyJ..."
export SENTRY_ORGANIZATION="my-org"
export SENTRY_PROJECT="my-project"

Example prompts:

1. Top errors:
"List the 10 most frequent errors in the past 7 days with stack traces"

2. Root cause analysis:
"The DatabaseConnectionTimeoutException occurs every 2 hours. What's the likely cause?"

3. Fix generation:
"Generate a code fix for the NullPointerException in PaymentProcessor.processRefund()"

4. Triage:
"Rank these errors by business impact: OutOfMemoryError, ValidationException, TimeoutError"

5. Pattern detection:
"Are there any correlated error patterns? (e.g., errors that happen together)"

Gotchas & Tips

Gotcha 1: Rate Limiting - Sentry API has rate limits (usually 25 req/sec for free tier) - Don't ask Claude to analyze 1000 errors at once - Solution: Filter by time range, severity, or tag

Gotcha 2: Incomplete Stack Traces - Sentry sometimes shows minified/obfuscated stack traces - Claude may not be able to pinpoint exact line of code - Solution: Ensure source maps are uploaded to Sentry

Gotcha 3: Missing Context - Stack trace alone isn't always enough - If Claude asks for: database logs, recent deployments, config changes—that's normal - In real scenarios, integrate logs from other sources (CloudWatch, etc.)

Tip 1: Tag Everything - Add context to Sentry errors: environment, feature flag, user segment - Claude can then filter by tag: "Errors only from feature flag 'new-checkout'"

Tip 2: Breadcrumbs - Sentry breadcrumbs show what happened before the error (API calls, state changes) - Claude uses breadcrumbs to understand error context better

Tip 3: Regression Detection - Ask Claude: "This error started appearing after we deployed version 2.3—what changed?" - Claude can correlate errors with deployment times

Tip 4: Environment Awareness - Errors in production vs. staging may have different root causes - Claude can analyze: "This only happens in production. Why?"

Lead-out

"You've seen how Claude becomes a debugging partner—reading Sentry errors and suggesting fixes faster than manual analysis. In the next video, we're staying in the communication layer but shifting to Slack—using Claude to read channel context and draft smart messages."

Reference URLs

  • Sentry MCP Server: https://github.com/getsentry/sentry-mcp
  • Sentry API Documentation: https://docs.sentry.io/api/
  • Sentry Auth Tokens: https://sentry.io/settings/account/api/auth-tokens/
  • Source Maps in Sentry: https://docs.sentry.io/platforms/javascript/sourcemaps/

Prep Reading

  • Create a Sentry test project with sample errors (5 min)
  • Generate auth token for MCP (2 min)
  • Review a few real error scenarios in your Sentry instance (5 min)

Notes for Daniel

  • Live data warning: If using production Sentry, sanitize error messages (remove customer data, API keys) before sharing screen.
  • Tone: Position Claude as a "second pair of eyes" for debugging, not a replacement for human understanding.
  • Failure scenario: If Sentry has few errors, prepare a pre-recorded demo or test project with common errors (NullPointer, Timeout, OutOfMemory).
  • Fun fact: Claude is particularly good at spotting patterns humans miss—correlating seemingly unrelated errors.
  • Next step momentum: Mention that once Claude suggests a fix, you can implement it in Claude Code (tying back to earlier modules).