Tracecat — Agentic Security Automation

Practical Notes

• GitHub: 3,598 stars · 359 forks; pushed 2026-05-12 (verified via GitHub API).
• Repo includes docker-compose.dev.yml + docker-compose.local.yml and a justfile wrapper for common workflows.
• Per README: sandboxed-by-default with nsjail and durable workflows on Temporal.

Main

A practical adoption path:

1. Start with one playbook. Pick a repeatable task (phishing triage, IOC enrichment, alert dedupe) and model it as a workflow.
2. Gate “risky tools.” Put human approval in front of actions that touch production systems or customer data.
3. Version your scripts. Use the code-native registry idea: keep custom Python tools in Git and sync them into Tracecat.
4. Treat MCP servers as dependencies. Prefer a small, curated set of MCP servers and pin their configs for reproducibility.

If you run this on developer laptops, keep secrets in env vars and rotate them often; for production, use a dedicated environment and audited credentials.

FAQ

Q: Do I need Kubernetes? A: No. The repo ships multiple Docker Compose files; Kubernetes is optional for production scaling.

Q: Can it run untrusted code safely? A: README says it uses nsjail sandboxes by default; still treat inputs as untrusted and apply least privilege.

Q: How do agents interact with it? A: Per README: you can build prompt-to-automations from your own agent harness, and tools can integrate with MCP servers.