Sentry Mcp

Sentry Toolkit

This repository contains the Sentry CLI, the Sentry MCP server, and shared packages. For the CLI, see the CLI guide and the published documentation at https://cli.sentry.dev/. The documentation index covers both products and shared contributor guidance.

The independent Docs MCP Worker provides documentation search and AI-generated answers backed by docs.sentry.io.

Sentry MCP

Sentry's MCP service is primarily designed for human-in-the-loop coding agents. Our tool selection and priorities are focused on developer workflows and debugging use cases, rather than providing a general-purpose MCP server for all Sentry functionality.

This remote MCP server acts as middleware to the upstream Sentry API, optimized for coding assistants like Cursor, Claude Code, and similar development tools. It's based on Cloudflare's work towards remote MCPs.

Getting Started

You'll find everything you need to know by visiting the deployed service in production:

https://mcp.sentry.dev

If you're looking to contribute, learn how it works, or to run this for self-hosted Sentry, continue below.

Claude Code Plugin

Install as a Claude Code plugin for automatic subagent delegation:

claude plugin marketplace add getsentry/sentry-mcp
claude plugin install sentry-mcp@sentry-mcp

This provides a sentry-mcp subagent that Claude automatically delegates to when you ask about Sentry errors, issues, traces, or performance.

For forward-looking tool variants and features:

claude plugin install sentry-mcp@sentry-mcp-experimental

Stdio vs Remote

While this repository is focused on acting as an MCP service, we also support a stdio transport. This is still a work in progress, but is the easiest way to adapt run the MCP against a self-hosted Sentry install.

Note: The AI-powered search tools (search_errors, search_traces, search_logs, search_issues, etc.) require an LLM provider (OpenAI, Azure OpenAI, Anthropic, or OpenRouter). These tools use natural language processing to translate queries into Sentry's query syntax. Without a configured provider, these specific tools will be unavailable, but all other tools will function normally.

The stdio transport ships as @sentry/mcp-server. Sign in once with npx @sentry/mcp-server@latest auth login, or set SENTRY_AUTH_TOKEN, which the server uses over a stored login. The login is shared with the sentry CLI, so an existing sentry auth login works too.

To use a token, create a User Auth Token in Sentry with the necessary scopes. As of writing this is:

org:read
project:read
project:write
team:read
team:write
event:write

Launch the transport:

npx @sentry/mcp-server@latest auth login   # once, to sign in
npx @sentry/mcp-server@latest              # uses that login
SENTRY_AUTH_TOKEN=sentry-user-token npx @sentry/mcp-server@latest

--access-token / SENTRY_ACCESS_TOKEN from older releases still work but are deprecated: they act as SENTRY_AUTH_TOKEN and print a warning on stderr. See docs/operations/stdio-auth.md.

Need to connect to a self-hosted deployment? Set SENTRY_HOST (e.g. SENTRY_HOST=sentry.example.com), which also scopes SENTRY_AUTH_TOKEN to that server. For isolated internal deployments that only expose plain HTTP, use an http:// SENTRY_URL instead.

Seer is not part of self-hosted Sentry, so the seer skill is left out of the default skill set whenever the host is not sentry.io. If your self-hosted deployment does run Seer, opt back in explicitly:

SENTRY_HOST=sentry.example.com npx @sentry/mcp-server@latest --skills=inspect,seer

You can also disable any other skill to prevent unsupported tools from being exposed:

SENTRY_HOST=sentry.example.com npx @sentry/mcp-server@latest --disable-skills=project-management

For self-hosted instances without TLS:

SENTRY_URL=http://sentry.internal:9000 npx @sentry/mcp-server@latest

Remote with an Explicit Sentry Token

Remote clients that support custom HTTP headers can pass an upstream Sentry API token directly to the Cloudflare transport:

{
  "mcpServers": {
    "sentry": {
      "url": "https://mcp.sentry.dev/mcp",
      "headers": {
        "Authorization": "Sentry-Bearer ${SENTRY_ACCESS_TOKEN}"
      }
    }
  }
}

Sentry-Bearer is intentionally separate from Bearer: Bearer is reserved for MCP OAuth access tokens. With Sentry-Bearer, the worker does not store, validate, exchange, or refresh the upstream token. It forwards the token through the same Sentry API calls used by OAuth-backed sessions, and the client or upstream provider remains responsible for token lifetime and refresh.

Direct remote auth defaults to all active MCP skills. You can narrow the exposed tools with ?skills=inspect,triage or ?disable-skills=seer.

Environment Variables

SENTRY_AUTH_TOKEN=           # Optional after `auth login`; wins over a stored login

# LLM Provider Configuration (required for AI-powered search tools)
EMBEDDED_AGENT_PROVIDER=     # Required when multiple provider keys are set: 'openai', 'azure-openai', 'anthropic', or 'openrouter'
OPENAI_API_KEY=              # Required if using OpenAI
ANTHROPIC_API_KEY=           # Required if using Anthropic
OPENROUTER_API_KEY=          # Required if using OpenRouter
OPENROUTER_MODEL=            # Optional OpenRouter model, defaults to 'openai/gpt-5.6-luna'
OPENROUTER_REASONING_EFFORT= # Optional OpenRouter reasoning effort, defaults to 'high'

# Optional overrides
SENTRY_HOST=                 # For self-hosted deployments (drops 'seer' from the default skills)
MCP_SKILLS=                  # Grant specific skills (comma-separated, e.g. 'inspect,seer')
MCP_DISABLE_SKILLS=          # Disable specific skills (comma-separated, e.g. 'project-management')

Important: Always set EMBEDDED_AGENT_PROVIDER to explicitly specify your LLM provider. Auto-detection based on API keys alone is deprecated and will be removed in a future release. See docs/operations/embedded-agents.md for detailed configuration options.

Example MCP Configuration

{
  "mcpServers": {
    "sentry": {
      "command": "npx",
      "args": ["@sentry/mcp-server@latest"],
      "env": {
        "SENTRY_AUTH_TOKEN": "your-token",
        "EMBEDDED_AGENT_PROVIDER": "openai",
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

If you leave the host variable unset, the CLI automatically targets the Sentry SaaS service. Only set the override when you operate self-hosted Sentry.

Setting SENTRY_HOST to a self-hosted host also drops the seer skill from the default set, since Seer is not available on self-hosted Sentry. For a self-hosted deployment that does run Seer, opt in with MCP_SKILLS:

{
  "mcpServers": {
    "sentry": {
      "command": "npx",
      "args": ["@sentry/mcp-server@latest"],
      "env": {
        "SENTRY_AUTH_TOKEN": "your-token",
        "SENTRY_HOST": "sentry.example.com",
        "MCP_SKILLS": "inspect,seer"
      }
    }
  }
}

Rate Limits

Tool calls are subject to the Sentry API's rate limits. The MCP server retries rate-limited requests, and transient Sentry server errors on read requests, after a short wait. When a limit lasts longer, the tool returns a Rate Limited message that says when to try again. Agents hit the per-second limits most often by running many Sentry tool calls in parallel, so run them one at a time if you see this message.

MCP Inspector

MCP includes an Inspector, to easily test the service:

pnpm inspector

Enter the MCP server URL (http://localhost:5173) and hit connect. This should trigger the authentication flow for you.

Note: If you have issues with your OAuth flow when accessing the inspector on 127.0.0.1, try using localhost instead by visiting http://localhost:6274.

Local Development

To contribute changes, you'll need to set up your local environment:

1. Set up environment and agent skills:

   make setup-env  # Creates .env files and installs shared agent skills
   

   This also runs npx @sentry/dotagents install to install shared skills from getsentry/skills into .agents/skills/ (symlinked into .claude/skills and .cursor/skills). If you need to update skills later, run it directly:

   npx @sentry/dotagents install
   

2. Create an OAuth App in Sentry (Settings => API => Applications):

   • Homepage URL: http://localhost:5173
   • Authorized Redirect URIs: http://localhost:5173/oauth/callback
   • Note your Client ID and generate a Client secret

3. Configure your credentials:

   • Edit .env in the root directory and add either OPENAI_API_KEY or OPENROUTER_API_KEY
   • Edit packages/mcp-cloudflare/.env and add:
     • SENTRY_CLIENT_ID=your_development_sentry_client_id
     • SENTRY_CLIENT_SECRET=your_development_sentry_client_secret
     • COOKIE_SECRET=my-super-secret-cookie

4. Start the development server:

   pnpm dev
   

Verify

Run the server locally to make it available at http://localhost:5173

pnpm dev

To test the local server, enter http://localhost:5173/mcp into Inspector and hit connect. Once you follow the prompts, you'll be able to "List Tools".

Tests

There are three test suites included: unit tests, evaluations, and manual testing.

Unit tests can be run using:

pnpm test

Evaluations require a .env file in the project root with some config:

# .env (in project root)
OPENAI_API_KEY=      # Use OpenAI-backed AI-powered tools
OPENROUTER_API_KEY=  # Or use OpenRouter-backed AI-powered tools

Note: The root .env file provides defaults for all packages. Individual packages can have their own .env files to override these defaults during development.

Once that's done you can run them using:

pnpm eval

Manual testing (preferred for testing MCP changes):

# Test with local dev server (default: http://localhost:5173)
pnpm -w run cli "who am I?"

# Test against production
pnpm -w run cli --mcp-host=https://mcp.sentry.dev "query"

# Test with local stdio mode (requires SENTRY_ACCESS_TOKEN)
pnpm -w run cli --access-token=TOKEN "query"

Note: The CLI defaults to http://localhost:5173. Override with --mcp-host or set MCP_URL environment variable.

Comprehensive testing playbooks:

• Stdio testing: See docs/testing/stdio.md for complete guide on building, running, and testing the stdio implementation (IDEs, MCP Inspector)
• Remote testing: See docs/testing/remote.md for complete guide on testing the remote server (OAuth, web UI, CLI client)

Development Notes

Automated Code Review

This repository uses automated code review tools (like Cursor BugBot) to help identify potential issues in pull requests. These tools provide helpful feedback and suggestions, but we do not recommend making these checks required as the accuracy is still evolving and can produce false positives.

The automated reviews should be treated as:

• ✅ Helpful suggestions to consider during code review
• ✅ Starting points for discussion and improvement
• ❌ Not blocking requirements for merging PRs
• ❌ Not replacements for human code review

When addressing automated feedback, focus on the underlying concerns rather than strictly following every suggestion.

Contributor Documentation

Looking to contribute? Start at the Toolkit documentation index for CLI, MCP, and shared guides. See AGENTS.md for repository rules and packages/cli/AGENTS.md for CLI-specific rules.