Sentry MCP server

Issues, stack traces and events from your Sentry organization.

How it connects

Runs on: Sentry's own MCP endpoint
Connects with: Your Sentry account, connected once for your org
Category: Developer Tools

Checked October 16, 2026 against Sentry's hosted server as listed on Oct 16, 2026, and the getsentry/sentry-mcp source for what execute_sentry_tool can reach

What it does

Sentry runs this server at https://mcp.sentry.dev/mcp for organizations on sentry.io, and you sign in with your Sentry account. An agent can search issues, errors, logs, traces and replays, open a stack trace from a link, resolve or assign issues, and ask Seer, Sentry's AI debugger, for a root cause. The code is open source in getsentry/sentry-mcp.

Tools

The hosted server lists 15 tools. Most searches accept plain English, and Sentry translates them into its own query syntax before running them.

Tool	What it does	Reads or changes
`find_organizations`	Lists the organizations you can access, with their slugs and region URLs.	Reads
`find_projects`	Lists or searches projects in an organization.	Reads
`search_issues`	Returns a list of grouped issues matching a query, such as unresolved errors in checkout.	Reads
`search_errors`	Searches error events and crashes, for counts, trends or single events.	Reads
`search_logs`	Searches application logs, including error and warning lines.	Reads
`search_traces`	Searches spans and traces for latency, throughput and slow requests.	Reads
`search_metrics`	Queries counters, gauges and distributions, such as p95 request duration.	Reads
`search_profiles`	Finds transaction and continuous profiles to inspect.	Reads
`search_replays`	Finds session replays, for example ones with rage clicks or errors.	Reads
`find_dropped_events`	Shows events Sentry dropped before storing them, by reason such as rate limits, quota or sampling.	Reads
`get_sentry_resource`	Fetches an issue, event, trace, span, replay, monitor or snapshot from a Sentry URL or an ID.	Reads
`search_sentry_tools`	Searches Sentry's larger catalog of tools that are not listed directly, and returns their schemas.	Reads
`execute_sentry_tool`	Runs any catalog tool by name, including ones that create, update or delete.	Changes
`update_issue`	Resolves, reopens, ignores, assigns or unassigns an issue.	Changes
`analyze_issue_with_seer`	Returns Seer's root cause analysis and suggested fix, and starts a new Seer run if the issue has none. A new run takes a few minutes.	Changes

Set it up

Sentry recommends scoping the URL to one project when you can: https://mcp.sentry.dev/mcp/<org-slug>/<project-slug>. The server then defaults every tool to that project and rejects requests for others. https://mcp.sentry.dev/mcp/<org-slug> scopes to an organization.

Claude Code, directly.

bash

claude mcp add --transport http sentry https://mcp.sentry.dev/mcp/<org-slug>/<project-slug>

Start a new session, run /mcp, and sign in to Sentry from there. How to add MCP servers to Claude Code covers scopes and the login flow.

Cursor, directly. In ~/.cursor/mcp.json:

json

{
  "mcpServers": {
    "sentry": {
      "url": "https://mcp.sentry.dev/mcp/<org-slug>/<project-slug>"
    }
  }
}

Cursor opens the Sentry sign-in the first time it connects. See How to set up MCP servers in Cursor for the rest.

Either way, Sentry shows an approval page that lists skills: Inspect Issues & Events, Seer, Triage Issues, and Manage Projects & Teams. All four are ticked the first time. The skills you leave ticked decide which tools the session gets, including what execute_sentry_tool can reach. Untick what the agent doesn't need before you approve.

Through Statio. Install Sentry from the catalog and press Connect. You approve access on Sentry's own page, and the connection belongs to your organization in Statio, so one person connects it for everyone. Then grant tools under Permissions. Nothing is granted until you do. The desktop app writes the client config, or by hand:

bash

claude mcp add --transport http sentry http://127.0.0.1:7777/sentry/mcp

In Cursor, use "url": "http://127.0.0.1:7777/sentry/mcp". The path uses the server's name in your dashboard. Agents that run unattended get their own credentials, as shown in Connect a client. If the server shows no tools, see How to debug an MCP server that won't connect.

What to hold back

execute_sentry_tool is the one to think about. Behind the 15 listed tools, Sentry keeps a catalog of about 60 more, and execute_sentry_tool runs any of them by name. With Manage Projects & Teams approved, that includes delete_alert_rule, delete_metric_monitor, delete_uptime_monitor, update_dsn, update_project, create_team and remove_team_from_project. A gateway that grants per tool sees only the name execute_sentry_tool, so granting it grants the whole catalog your Sentry skills allow. Most of the reads an agent needs are covered by get_sentry_resource and the search_* tools. If you do grant execute_sentry_tool, untick Triage Issues and Manage Projects & Teams on Sentry's approval page, so the catalog behind it can only read, apart from Seer.

The damage from the project-management writes is slow to show. update_dsn can deactivate a client key or put a rate limit on it, and your app stops reporting errors to Sentry. Deleting an alert rule means the on-call engineer never hears about the next outage. Changing a project's slug with update_project breaks scripts and configs that refer to the old slug.

update_issue for a coding agent. A coding agent that fixes a bug may want to mark the issue resolved. If the fix didn't work, the issue stays closed until it shows up again as a regression. Ignoring an issue silences its alerts for as long as the agent chose. Assigning sends a notification to a real person. Let the agent link the issue in its pull request and leave the status to a human, or to Sentry's resolve-on-release.

analyze_issue_with_seer costs money. When an issue has no Seer analysis yet, the tool starts one in Sentry. Seer is a paid add-on to your Sentry subscription. Current plans bill it per active contributor, and Sentry's older Seer pricing charged $1.00 for each Issue Fix run. An agent that calls it on every issue it reads can run up a bill. Grant it only to agents a person drives, and drop the Seer skill on the approval page for the rest.

A triage bot usually needs the search_* tools, get_sentry_resource and maybe update_issue for assignment. It doesn't need execute_sentry_tool.

For how per-tool grants work and how a refused call is recorded, see MCP permissions.

Questions

Yes. Sentry hosts it at https://mcp.sentry.dev/mcp and publishes the code as getsentry/sentry-mcp on GitHub. It signs in with OAuth, so there is no token to create.

Yes. Connect to https://mcp.sentry.dev/mcp/your-org/your-project. Sentry ties the session to that project and refuses requests for other projects with the same token.

execute_sentry_tool runs tools from a larger catalog that search_sentry_tools finds. Which of those it can reach depends on the skills you approved when you signed in.

Sentry points self-hosted users to the stdio package, @sentry/mcp-server, run with a user auth token and --host set to your Sentry host. Seer is left out by default there, because self-hosted Sentry doesn't include it.

What Statio adds

You choose which of its tools each agent may call. A tool you hold back is removed from that agent's tool list, and a call to it is refused.
Agents get a Statio token, never the Sentry one.
Every call is recorded with the agent that made it, when, how long it took and whether it succeeded.