Let Seer write contained fixes itself, and hand its root cause and plan to a Claude, Cursor, or GitHub Copilot agent when a fix needs a running environment or spans repositories. When work starts in the editor or terminal, connect Sentry to the agent your engineers already run. Every path should end in a pull request that a person approves.
When you need this
- Your engineers already use Claude Code, Cursor, or Copilot and ask what Seer adds.
- Seer's own fixes need rounds of iteration that an agent with a running environment can do.
- You are deciding between Seer and a triage pipeline built on your own agent.
Evaluation criteria
Seer writes the fix
When should Seer open the pull request itself?
Look forUse Seer for contained fixes in GitHub.com or GitLab.com repositories. It drafts a change and opens a pull request. On GitHub, it also responds to failing CI checks and review comments.
Hand off to a cloud agent
When should a coding agent implement Seer's plan?
Look forHand off when the fix needs a running environment or spans repositories, or when your team routes work through one agent. Seer supplies the root cause and plan. Claude Agent runs as a managed agent on your Anthropic API key. Cursor Cloud Agents use your Cursor API key. Copilot runs in GitHub Actions. Handoff requires GitHub. Set handoff on each project, because in July 2026 Wamiz's CTO found that the organization-wide setting did not persist and the per-project setting did.
Bring Sentry into your agent
When should your own agent pull from Sentry?
Look forUse this path when work starts in the editor or terminal. Claude Code connects with claude mcp add --transport http sentry https://mcp.sentry.dev/mcp. Without Seer, the agent reads raw events. With Seer enabled, it can request Seer's root cause analysis. In September 2026, Shepherd's maintainers replaced Seer with their own fix pipeline, because Seer is closed source, costs US$40 per active contributor, and knows nothing about their house rules.
Treat Seer's output as data
How does the agent use Seer's analysis?
Look forHave the agent verify Seer's analysis as a hypothesis before acting on it. Liz Tom's Copilot extension places Seer's root cause and plan in an untrusted data block. Under CVE-2026-90999, a fabricated event can shape the analysis an agent receives.
Keep one review gate
Where does a person decide?
Look forEvery path ends in a branch or pull request. Keep human approval on it, and cap how many pull requests an automated run can open.
Questions to ask vendors
- Where do our bug fixes start today: Sentry, Slack, the tracker, or the editor?
- Which coding agent do we already pay for, and does it have network and repository access for Seer's handoff?
- Do we need Seer's diagnosis, or is raw Sentry data enough for most of our issues?
- Which of our agents can install packages while it works on a Seer handoff?
Red flags
- Paying for Seer and a separate agent pipeline that both run on every issue.
- An agent prompt that treats Seer's analysis as instructions.
- Automatic handoff configured at the organization level without checking each project.
- Agents with write access to repositories they were never asked to change.
Products in this piece
Sources
- Official documentation
- 7
- Public source
- 5
- Sentry docs: Autofix
- Sentry docs: Coding Agents
- Sentry docs: Claude Agent
- Sentry docs: Cursor Agent
- Sentry docs: GitHub Copilot Agent
- Sentry blog: Seer, the Sentry MCP and CLI, or your own coding agent (Oct 2, 2026)
- Claude Code docs: Connect Claude Code to tools via MCP
- getsentry/sentry issue #120647: Seer handoff to agents not persisting (Jul 27, 2026)
- Liz Tom, awesome-copilot pull request #8: seed the Copilot fix with Seer's analysis (Sep 11, 2026)
- erwins-enkel Shepherd issue #2467: Sentry auto-fix as a Seer replacement (Sep 24, 2026)
- jblake on Hacker News: Seer is interesting but unnecessary next to Claude (Apr 19, 2026)
- CERT/CC Vulnerability Note VU#212479: Sentry Seer (Sep 16, 2026)