Pentest-AI
Field notes

MCP and the CLI work together

Your existing AI client can drive the investigation. The CLI remains useful for repeatable workflows.

Updated 12 September 2026

Keep the client you already use

An MCP connection lets your AI client call Pentest-AI's tools. You can ask it to start an authorized engagement, inspect findings or request verification without switching to a different hosted product.

Codex and Claude Code can both use the local ptai mcp server. The client provides model access and its own tool-approval experience.

The terminal still matters

The standalone CLI is useful when you want explicit commands, saved output or a CI workflow. It can use a cloud model API, a local Ollama model or deterministic checks without a model.

MCP and the CLI are complementary interfaces. Neither interface changes the meaning of an oracle-verified finding.

Let evidence settle the claim

The model helps investigate. An oracle independently checks whether the exploit reproduces and whether a control behaves differently. Keeping those roles separate makes the result more useful to the person fixing the application.

Connect your AI client →

Found something unclear?Open an issue ↗