How to use the Sentry MCP server for error tracking

Log into sentry.io and go to Settings > Developer Settings > Auth Tokens (or Settings > Account > API > Auth Tokens for personal tokens). Create a new token with the scopes: project:read, org:read, event:read, and issue:read at minimum. If you want the AI to resolve or assign issues, also add issue:write. Copy the token — Sentry will only show it once.

1Required token scopes:
2- org:read — List organizations and teams
3- project:read — List projects
4- issue:read — Search and view issues
5- event:read — View error events and stack traces
6
7Optional scopes:
8- issue:write — Resolve, assign, or update issues
9- member:read — View team member information

Open your MCP host's configuration file and add a sentry entry. The Sentry auth token is passed as an environment variable. The configuration uses npx with the -y flag for automatic installation.

1{
2  "mcpServers": {
3    "sentry": {
4      "command": "npx",
5      "args": [
6        "-y",
7        "@sentry/mcp-server-sentry"
8      ],
9      "env": {
10        "SENTRY_AUTH_TOKEN": "sntrys_your_token_here"
11      }
12    }
13  }
14}

Fully quit and reopen Claude Desktop, Cursor, or VS Code. The Sentry MCP server will authenticate with your account and expose tools for searching issues, viewing events, and listing projects. Verify the connection by checking the tools list.

Start by asking the AI to list your Sentry projects and show recent issues. This confirms the connection works and gives you an overview of your error landscape. The AI can filter issues by status, frequency, first/last seen dates, and assigned team.

1Example prompts:
2
3"List all my Sentry projects and their error counts"
4
5"Show me the top 10 unresolved issues in the frontend-app project, sorted by frequency"
6
7"Search for issues related to 'TypeError' in the last 7 days"

The real power comes from deep error analysis. Ask the AI to examine specific issues — view the stack trace, read the breadcrumbs (user actions leading to the error), check tags and context, and look at affected users. The AI can then suggest fixes based on the error data, especially when combined with your codebase through the filesystem MCP server.

1Example prompts:
2
3"Show me the full stack trace for issue PROJ-1234 and explain what's causing it"
4
5"What are the breadcrumbs leading up to the most frequent error in my project?"
6
7"Analyze the latest 5 events for the 'Cannot read property of undefined' issue and find the common pattern"

For VS Code with GitHub Copilot, use the servers key instead of mcpServers. Create a .vscode/mcp.json file in your workspace root.

1{
2  "servers": {
3    "sentry": {
4      "command": "npx",
5      "args": [
6        "-y",
7        "@sentry/mcp-server-sentry"
8      ],
9      "env": {
10        "SENTRY_AUTH_TOKEN": "sntrys_your_token_here"
11      }
12    }
13  }
14}