Skip to main content

Overview

OpenSRE connects to Sentry’s hosted Model Context Protocol (MCP) server. That exposes Sentry products — issues, events, traces, replays, releases, monitors, and Seer AI root-cause analysis — as tools the agent can call during an investigation. This is distinct from the Sentry issue integration, which is a narrower REST client for issue and event lookup. Use MCP when you want the agent to explore Sentry data directly, including Seer-assisted debugging. “Hosted MCP” here means Sentry’s mcp.sentry.dev endpoint — not OpenSRE Cloud.

Prerequisites

  • A Sentry account (hosted SaaS, or self-hosted via the SENTRY_MCP_HOST override)
  • A Sentry user auth token. Required scopes depend on the skills you use:
    • org:read — read-only inspection of issues, events, traces, releases, monitors (default)
    • event:write, team:write, project:write — for triage and project-management skills

Setup

Option 1: Interactive CLI

Select Sentry (MCP) when prompted, then paste your user auth token. Setup uses the hosted Streamable HTTP transport; keep the default URL unless you run self-hosted Sentry. To run a local server instead, set SENTRY_MCP_MODE=stdio via environment variables.

Option 2: Environment variables

To run a local Sentry MCP server instead of the hosted endpoint:
The search_events and search_issues tools translate natural-language queries and require an OPENAI_API_KEY in the MCP server’s environment. The rest of the MCP server works without it.

Option 3: Persistent store

Credentials

Create a Sentry user auth token in account settings with at least org:read for read-only investigation. Add write scopes only if you enable triage or project-management skills.

Investigation tools

Typical flow: call list_sentry_tools first, then call_sentry_tool with the chosen name and arguments.

Verify

A successful check connects to the MCP server and reports how many tools it discovered (detail shape: Sentry MCP connected via {mode} ({endpoint}); discovered N tool(s).).

Troubleshooting

Security

  • Prefer org:read only unless you need write skills.
  • Store SENTRY_MCP_AUTH_TOKEN in .env or your secret manager — not in source control.
  • Scope with SENTRY_MCP_ORGANIZATION_SLUG / SENTRY_MCP_PROJECT_SLUG when possible.