Does Sentry have an MCP server, and what can it do?
Yes. Sentry runs a hosted MCP server at https://mcp.sentry.dev/mcp with OAuth sign-in, and publishes a stdio package, @sentry/mcp-server, for self-hosted Sentry. Coding agents can use it to search issues, errors, logs and traces, ask Seer for analysis and, if you grant the triage skill, resolve and assign issues.
What it can do
Sentry groups the tools into skills you grant when you authorize. These names and defaults are from its open-source repository on 2026-11-06:
| Skill | Covers | Default |
|---|---|---|
| Inspect Issues & Events | Read-only: issues, events, traces, replays, releases, monitors, profiles, documentation, project metadata | On |
| Seer | Sentry's AI debugger, to analyze and root-cause issues | On |
| Triage Issues | Resolve, assign and update issues | Off |
| Manage Projects & Teams | Create and change projects, teams, DSNs, monitors and alert rules | Off |
The main tools include search_issues, search_errors, search_logs, search_traces, search_replays, update_issue and analyze_issue_with_seer. If you want the agent to mark issues resolved after a fix, turn on Triage when you sign in.
How to connect it
Any MCP client that supports HTTP can use the hosted URL. In Claude Code:
claude mcp add --transport http sentry https://mcp.sentry.dev/mcp
# then run /mcp inside Claude Code and sign inSentry lets you scope a connection to an organization or project and recommends a project where possible. Scoped tools default to that project, and discovery tools you no longer need are hidden.
Limits to know
- Sentry describes the service as "primarily designed for human-in-the-loop coding agents", not a general-purpose server for all Sentry functionality.
- With the stdio package, the AI-powered search tools (search_errors, search_issues, search_logs, search_traces and others) need your own LLM provider key. Without one, those tools are unavailable and the rest still work.
- Seer needs a subscription, according to Sentry's pricing page, and is not part of self-hosted Sentry, so that skill is left out of the default set there.
- The server reads what your Sentry SDKs already send, so your app has to be reporting to Sentry first.
Which server fits
If your errors are already in Sentry, use Sentry's server. It covers tracing, session replay and profiling, and Sentry has browser and mobile SDKs, which Runheed does not.
Runheed is a smaller tool built around the agent loop: errors only, Node.js and Python SDKs plus an HTTP endpoint, nine tools, and resolve available to the agent from the first sign-in. Agent name, session and tool are named fields on each error. Both SDKs run in one process, so you can keep Sentry and add Runheed for the agent. See Runheed vs Sentry for the full comparison, and error tracking MCP servers compared for other vendors.
Sources
- mcp.sentry.dev: URL, OAuth, scoping
- getsentry/sentry-mcp on GitHub: stdio package, skills, tool names, LLM requirement
- Claude Code MCP docs: the add command
- Sentry pricing: Seer subscription
Tool lists change. Check Sentry's pages before you rely on a detail here.
Questions
- What is the Sentry MCP server URL?
- https://mcp.sentry.dev/mcp. All connections to the hosted server use OAuth, unless your client sends a Sentry API token in a custom header.
- Can the Sentry MCP server resolve issues?
- Yes, through update_issue, part of the Triage skill. That skill is off by default in the OAuth sign-in, so turn it on when you authorize.
- Does the Sentry MCP server work with self-hosted Sentry?
- Yes, through the stdio package with SENTRY_HOST set. Sentry calls stdio a work in progress. Seer is left out by default because it is not part of self-hosted Sentry.
- Can I run Sentry and another error tracker together?
- Yes. Runheed's SDK and Sentry's run in the same process, and your agent can have both servers connected.