Sentry MCP Server

npx @sentry/mcp-server@latest --access-token=YOUR_TOKEN

Sentry MCP Server

The Sentry MCP Server is an official integration from Sentry that exposes error monitoring, issue tracking, and performance data as MCP tools consumable by any AI coding assistant. It operates as a remote streamable-HTTP server hosted at mcp.sentry.dev, so no local binary installation is required for the default setup. As of early 2026, it covers 8+ tools across issue management, event retrieval, and Seer AI analysis.

Sentry holds error data from millions of production deployments. With the MCP server, that data is addressable inside a Claude session: describe a crash, Claude calls find_issues, inspects the stack trace via get_event_for_issue, and proposes a fix — no copying from the Sentry UI.

For broader devops-and-infra workflows, combining Sentry MCP with the GitHub MCP server closes the full debug-to-PR loop: Sentry surfaces the issue, GitHub creates the branch and opens the fix.

How does it work?

The server acts as authenticated middleware between Claude and the Sentry REST API. You supply a SENTRY_ACCESS_TOKEN either as an environment variable or via the --access-token CLI flag. Claude then calls tools in sequence: first resolving the project slug, then fetching matching issues, then retrieving the raw event payload.

Remote vs. stdio transport

The production path connects over streamable-HTTP to mcp.sentry.dev using OAuth — no local process needed. The stdio path (npx @sentry/mcp-server@latest) is available for self-hosted Sentry instances where the remote service cannot reach your intranet deployment.

AI-powered search tools

Tools like search_events translate natural-language queries into Sentry’s query syntax using an LLM provider (OpenAI or Anthropic). These tools require a second API key alongside the Sentry token. If you omit it, the other tools still function — only the NLP-to-query translation is disabled.

Tool surface

Tool	What it does
find_issues	Search issues by query, project, or date range
get_issue_details	Fetch full issue metadata including tags and assignee
get_event_for_issue	Retrieve raw event payload with stack trace
get_seer_issue_summary	Pull Seer’s AI-generated root cause analysis
search_events	NLP query to Sentry discover query (requires LLM key)
list_projects	Enumerate projects in an org
update_issue	Change status, assignee, or priority
create_team	Create a new Sentry team in the org

When to use it

Use the Sentry MCP server when you want Claude to investigate real production errors without manual copy-paste from the Sentry dashboard. It works best for on-call debugging sessions, writing regression tests from a known crash event, and triage workflows where you want Claude to sort and assign issues before a sprint starts.

Teams that already use the GitHub MCP server for repository operations and need error context alongside commit history get the highest leverage from pairing both servers in a single Claude Code session.

Frequently asked questions

Does the Sentry MCP server support Sentry SaaS only, or self-hosted too?

Both. The remote server at mcp.sentry.dev targets Sentry SaaS. For self-hosted installations, run the stdio transport locally with npx @sentry/mcp-server@latest --access-token=TOKEN --host=sentry.yourcompany.com. Append --insecure-http for HTTP-only internal deployments.

Which transport should I choose — streamable-HTTP or stdio?

Use streamable-HTTP (the default remote server) for Sentry SaaS: nothing to install, OAuth handles auth. Use stdio when you need to connect to a self-hosted Sentry instance the remote service cannot reach, or when you want to pin a specific server version for a reproducible environment.

Are AI-powered search tools available on all plans?

The AI tools (search_events and similar) require a separately configured LLM provider key (OpenAI or Anthropic) in addition to your Sentry auth token. The feature works on any Sentry plan but requires you to supply and pay for the LLM provider separately.