MCP Server
Sentry MCP
Remote MCP server acting as middleware to the Sentry API, optimized for coding assistants to debug errors, analyze issues, and search traces
What is Sentry MCP?
Sentry MCP is an official server designed for human-in-the-loop coding agent workflows, focusing on developer debugging use cases rather than providing comprehensive Sentry API coverage. Deployed as a remote service at mcp.sentry.dev/mcp, the server acts as middleware between MCP clients and the upstream Sentry API, implementing tool selection and priorities optimized for Cursor, Claude Code, and similar development environments. The server supports both remote OAuth-based authentication for the hosted service and stdio transport for self-hosted Sentry deployments. AI-powered search tools including search_events, search_issues, and related capabilities require an LLM provider configuration (OpenAI, Azure OpenAI, Anthropic, or OpenRouter) to translate natural language queries into Sentry's query syntax. The stdio mode requires a Sentry user auth token with org:read, project:read, project:write, team:read, team:write, and event:write scopes. Self-hosted deployments can use the --host flag to connect to internal Sentry instances, with --insecure-http for non-TLS environments.
Key capabilities
AI-Powered Natural Language Search with LLM Provider Integration
Implements search_events, search_issues, and related tools that use embedded LLM agents (OpenAI, Anthropic, Azure OpenAI, or OpenRouter) to translate natural language queries into Sentry's query syntax, requiring EMBEDDED_AGENT_PROVIDER configuration.
Dual Transport Support for Remote and Self-Hosted Deployments
Operates as a remote MCP server at https://mcp.sentry.dev/mcp with OAuth authentication for SaaS Sentry, or runs in stdio mode via npx @sentry/mcp-server for self-hosted instances with SENTRY_ACCESS_TOKEN environment variable.
Skill-Based Tool Organization with Disable Flags
Groups tools into skills like inspect, triage, and seer, with --disable-skills command-line flag to prevent unsupported tools from being exposed (e.g., --disable-skills=seer for self-hosted instances without Seer support).
Claude Code Plugin with Automatic Subagent Delegation
Installable as a Claude Code plugin via claude plugin install sentry-mcp@sentry-mcp, providing automatic delegation to a sentry-mcp subagent when users ask about Sentry errors, issues, traces, or performance.
Use cases
Error Root Cause Analysis in Development Workflows
Developers debugging production issues can ask AI assistants to search Sentry issues by natural language description, retrieve event context including stack traces and breadcrumbs, and analyze patterns across related events without manual dashboard filtering.
Self-Hosted Sentry Integration for Internal Deployments
Organizations running self-hosted Sentry can deploy the stdio transport with --host=sentry.internal and --insecure-http flags, using user auth tokens instead of OAuth to enable AI assistants to access internal error tracking data.
Issue Triage with AI-Assisted Priority Assessment
AI coding assistants can use search and retrieval tools to identify high-impact errors, group related issues, assess frequency and user impact, and suggest priority levels based on trace data and event volume patterns.
Multi-Provider LLM Flexibility for Enterprise Policies
Organizations with specific LLM provider requirements can configure EMBEDDED_AGENT_PROVIDER to use internal Azure OpenAI endpoints, OpenRouter proxies, or Anthropic Claude, avoiding dependency on a single AI service for search translation.
MCP capabilities
Tools
search_issues
search_events
Inspect issue details
Manage project context
Connection and auth
- Sentry user auth token with org:read, project:read, project:write, team:read, team:write, event:write scopes
- EMBEDDED_AGENT_PROVIDER environment variable set to openai, azure-openai, anthropic, or openrouter for AI search
- Corresponding API key (OPENAI_API_KEY, ANTHROPIC_API_KEY, OPENROUTER_API_KEY) for LLM provider
- Optional SENTRY_HOST for self-hosted deployments
Client configuration
Claude Code
{
"type": "http",
"url": "https://mcp.sentry.dev/mcp"
}