SOAR that assumes your automation is an agent, not a runbook

What it does

Tracecat combines agents, low-code workflows, case management, and lookup tables into one self-hostable platform aimed at security teams. You can drive it from your own agent harness via its MCP server, or build workflows in a visual builder with if-conditions and loops. It ships with 100+ pre-built connectors to enterprise tools over HTTP, SMTP, gRPC, and OAuth, plus a custom registry that turns your own Python scripts into agent tools and workflow steps.

The interesting bit

The “prompt-to-automations” pitch is the angle: instead of clicking through a GUI, you let an agent harness like Claude Code or Codex assemble the automation for you. Underneath, untrusted code and agents run in nsjail sandboxes and workflows execute durably on Temporal — the boring infrastructure choices are the actual security story.

Key highlights

• Sandboxed-by-default execution via nsjail or pid runtimes
• Durable workflow execution on Temporal; PostgreSQL and S3-compatible storage underneath
• MCP server support, both remote HTTP/OAuth and local via npx/uvx
• SAML/OIDC in the open-source edition — no SSO paywall
• Exportable audit logs for piping into your SIEM

Caveats

• AGPL-3.0, with a paid Enterprise Edition directory (packages/tracecat-ee) and a PolyForm-licensed Kubernetes deployment submodule — the Helm chart is a private OCI artifact, so self-hosting on k8s is murkier than it sounds
• Actively developed; the README itself tells you to check the changelog before updating
• Enterprise features like RBAC/ABAC, human-in-the-loop approvals, and workflow version control are not in the OSS edition

Verdict

Worth a look if you run a security team and want SOAR-style automation you can self-host and script from an agent. If you need fine-grained access control or k8s deployment out of the box, expect to talk to sales.