Give coding agents Sentry evidence, not blanket control
Production debugging is faster when an agent can inspect the real issue, events, project, and trace. AIP routes Sentry's official MCP tools into each coding agent while keeping triage changes and long-tail operations behind a policy you can review.
What your agent can do
Direct read tools cover organizations, projects, issues, events, resources, and Seer analysis. Issue updates are explicit writes. Catalog execution is marked write-capable here because the selected long-tail operation may mutate Sentry even though many catalog entries are reads.
| Tool | What it does | Access |
|---|---|---|
| find_organizations | List or resolve Sentry organizations available to the authenticated user. | read |
| find_projects | Find projects within an accessible organization. | read |
| search_issues | Search issue data with project, environment, time, and natural-language criteria. | read |
| search_events | Search errors, spans, transactions, or other event data within the active Sentry scope. | read |
| get_sentry_resource | Retrieve a Sentry resource from its URL or identifier with structured context. | read |
| analyze_issue_with_seer | Ask Sentry Seer for root-cause analysis and fix recommendations when the account supports it. | read |
| search_sentry_tools | Discover available long-tail Sentry operations and their executable input schemas. | read |
| execute_sentry_tool | Run a catalog operation found through search; the selected operation can read or mutate data. | write |
| update_issue | Change issue assignment, status, or other supported triage fields. | write |
Sentry exposes a compact direct surface plus a searchable catalog of long-tail operations. The visible set varies with Sentry skills, account capabilities, and organization or project constraints.
Capability snapshot: 1 September 2026 · provider source
Bring the incident facts to the code
Without Sentry access, a coding agent works from a pasted stack trace that may omit the release, environment, tags, breadcrumbs, and neighboring events. With read tools, it can inspect the issue in place, compare occurrences, follow a resource URL, and connect the failing frame to the repository. That is enough for most diagnosis; resolving or assigning the issue is a separate decision.
Use the AIP control room to leave searches on while putting
update_issue and catalog execution behind Confirm. Combine Sentry with the
other systems on the integration hub when incident evidence
needs a Linear issue or a PostHog product-impact check.
Connect the official Sentry MCP
- Add Sentry from the AIP dashboard and finish the hosted Sentry OAuth flow for the intended organization access.
- Scope an agent to the direct tools it needs. If a workflow uses the Sentry tool catalog, expose search first and keep the generic executor supervised.
- Configure the AIP agent URL in the coding client. The same Sentry connection then works from Claude Code, Codex, Cursor, or another MCP-capable tool.
The hosted Sentry server decides which operations are available for the user, features, and active constraints. AIP keeps the credential centralized and adds the client-facing policy layer described in How it works.
Close the loop without auto-closing the issue
The most useful workflow reads Sentry, inspects code, runs a focused test, and returns to Sentry with a proposed owner or resolution. Granting the final update can save a manual step, but it should follow evidence such as a merged fix or a verified release rather than the agent’s confidence alone. Confirm mode keeps that last state change visible.
If different repositories map to different Sentry organizations or projects, separate agents and connections are easier to reason about than one global debugger. The pricing page helps size that split without weakening the boundary between production areas.
What to ask your agent
A good Sentry prompt identifies the service, environment, and time window, then asks the agent to connect evidence to code before changing issue state: