Sentry MCP

Sentry MCP is an open-source Model Context Protocol server built and maintained by Sentry (Functional Software, Inc.) that bridges Sentry's error monitoring and observability platform with LLM-powered coding agents. Currently in beta, it is deployed as a remote MCP service at mcp.sentry.dev and is designed primarily for human-in-the-loop developer workflows rather than general-purpose automation.

What It Is

Sentry MCP acts as middleware between Sentry's upstream API and AI coding assistants such as Cursor, Claude Code, Codex, Amp, Gemini CLI, OpenCode, Warp, Windsurf, and Zed. Developers can paste a Sentry issue URL directly into their coding agent and ask it to investigate or fix the problem — the MCP server fetches issue details, invokes Seer (Sentry's AI root-cause analysis engine) to analyze stack traces, and returns structured context that the LLM uses to propose and apply code fixes. The project is written in TypeScript, hosted on GitHub under the getsentry organization, and has accumulated over 800 stars since its creation in March 2025.

Core Capabilities

The server exposes a set of MCP tools organized around three main workflows:

• Fix Bugs — Fetch issue details, analyze stack traces with Seer, and let the coding agent apply a fix directly to the codebase, then validate with automated tests.
• Instrument Your App — Query traces, metrics, and errors to understand application behavior and guide instrumentation decisions.
• Search Things — Use AI-powered natural language search (backed by OpenAI, Anthropic, Azure OpenAI, or OpenRouter) to translate queries into Sentry's query syntax and find issues, events, or spans across projects.

AI-powered search tools (search_events, search_issues, etc.) require a configured LLM provider; all other tools function without one.

Deployment Model

Sentry MCP supports two transport modes:

• Remote (Streamable HTTP/SSE) — The production service at https://mcp.sentry.dev/mcp uses OAuth for authentication and runs on Cloudflare Workers. This is the recommended path for most users and requires no local setup beyond adding the server URL to a supported coding agent.
• Stdio — A local npx @sentry/mcp-server@latest command supports self-hosted Sentry deployments. Users supply a Sentry User Auth Token with scopes for org, project, team, and event access. Self-hosted instances can disable specific skills (e.g., --disable-skills=seer) when features like Seer are unavailable.

Path constraints on the remote URL (/:organization or /:organization/:project) let teams scope the MCP session to a specific org or project, automatically hiding irrelevant tools.

Integration Surface

The server integrates with a wide range of AI coding environments through standard MCP configuration:

• Claude Code (including as a Claude Code plugin via claude plugin marketplace add getsentry/sentry-mcp)
• Cursor, VSCode, Windsurf, Zed
• Codex, Amp, Gemini CLI, OpenCode, Warp

A live demo on the homepage lets users connect their Sentry account via OAuth and test MCP tool calls against real project data in a browser chat interface.

Update: Version 0.37.0

The latest release is 0.37.0, published on July 2, 2026. The repository shows active development with the last push on July 31, 2026. Recent commits visible in the homepage demo include schema improvements for trace responses — specifically adding a TraceIssueSchema to handle standalone error objects returned alongside span data in Sentry's trace API, and filtering logic to correctly separate span and issue items during trace processing. The README notes that auto-detection of LLM providers based on API keys is deprecated and will be removed in a future release, with EMBEDDED_AGENT_PROVIDER now required when multiple provider keys are set.