Sentry MCP Server: Debug Production Errors with AI

Production is on fire. Your on-call rotation kicks in at 2 AM and you need context fast. Instead of switching between terminals, Slack, and the Sentry dashboard, the Sentry MCP server lets you ask Claude directly: "What are the top unresolved errors in the last hour?" — and get a full stack trace, frequency trend, and debugging hypothesis in one response.

What Is the Sentry MCP Server?

The Sentry MCP server is an MCP server that wraps the Sentry REST API and exposes it as Claude-callable tools. Once configured, Claude can query your Sentry organization's projects, issues, events, and statistics using plain language — no need to remember Sentry's query syntax or navigate the UI.

The server is published as @modelcontextprotocol/server-sentry on npm and is maintained as part of Anthropic's official MCP server collection.

What the Sentry MCP Server Can Do

The server exposes the following tools to Claude:

• list_projects — returns all projects in your Sentry organization
• list_issues — query issues with filters: unresolved status, by tag, date range
• get_issue — full issue details including the complete stack trace
• resolve_issue — mark an issue as resolved
• list_events — list individual events for a given issue
• get_project_stats — error counts and trends for a project over time
• search_issues — Sentry search query passthrough for power users

Prerequisites

Before installing, make sure you have:

• A Sentry account — the free Developer plan covers 5,000 errors/month, which is enough to get started
• A Sentry Auth Token — navigate to User Settings → Auth Tokens → Create New Token, and enable the project:read scope (add project:write if you want Claude to resolve issues)
• Your Sentry organization slug — visible in the URL when you log in: sentry.io/organizations/YOUR-ORG-SLUG/
• Node.js 18+ installed — run node --version to check

How to Install the Sentry MCP Server

Installation follows the standard MCP server installation pattern — no separate install step is required because npx handles it at runtime.

Step 1: Open Your Claude Desktop Config

Open the Claude Desktop configuration file. On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows: %APPDATA%\Claude\claude_desktop_config.json. See the Claude Desktop config JSON guide for full details.

Step 2: Add the Sentry Server Block

Add the following to your config file. If you already have other MCP servers, add this block inside the existing mcpServers object.

{
  "mcpServers": {
    "sentry": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-sentry"],
      "env": {
        "SENTRY_AUTH_TOKEN": "sntryu_your_token_here",
        "SENTRY_ORG": "your-org-slug"
      }
    }
  }
}

Step 3: Restart Claude Desktop

Save the file and fully quit and reopen Claude Desktop. You should see a hammer icon or tool indicator confirming the Sentry MCP server is connected. Ask Claude "list my Sentry projects" to verify the connection.

Real-World Use Cases

Here is what you can ask Claude once the Sentry MCP server is running:

• "Show me the 5 most frequent unresolved errors in production this week"
• "Get the full stack trace for issue PROJ-123"
• "How many errors occurred in the last 24 hours in the checkout service?"
• "Which errors are new since yesterday's deploy?"
• "Mark issue PROJ-456 as resolved"

The most powerful use case is on-call triage: paste a Sentry issue ID into Claude, ask for the stack trace, and immediately ask "what is likely causing this and how do I fix it?" — Claude uses the trace context to give you a targeted hypothesis rather than generic advice.

Pros of the Sentry MCP Server

• Stack trace plus Claude equals instant debugging help — the combination is genuinely faster than reading traces manually
• Excellent for on-call when you need fast context without switching between five browser tabs
• Error trend analysis in natural language — ask "is this error getting worse?" and get a real answer
• Works with any Sentry plan including the free tier
• Official Anthropic-maintained package — reliable and updated alongside the MCP spec

Cons of the Sentry MCP Server

• Free tier limits history depth — you can only query events within the retention window (90 days on paid plans, 30 days free)
• Resolving issues from Claude bypasses your team's review process — use this carefully in team environments and consider keeping project:write scope off by default
• Large stack traces can fill Claude's context window, especially for deeply nested traces in Java or .NET applications
• No support for Sentry's advanced features like Replays, Profiling, or Crons through MCP