pentestagent

PentestAgent is a Python framework that puts Claude (or any LiteLLM-compatible model, including GPT-5) at the center of black-box penetration testing workflows. Users interact through a terminal UI with four distinct modes: Assist for single-shot instructions, Agent for autonomous task execution, Crew for orchestrated multi-agent runs, and Interact for guided conversational testing. Built-in tools cover terminal execution, a Playwright-based browser, notes, web search via Tavily, and a standout self-spawning mechanism called spawn_mcp_agent, which lets a running agent dynamically create isolated child agents as subordinate MCP servers over stdio, enabling hierarchical delegation without external orchestration. The framework connects to Claude through the Anthropic API using models such as claude-sonnet-4-20250514, and it exposes an MCP interface so additional servers can be attached at runtime via the /mcp command. A Kali Linux Docker image ships with Metasploit, sqlmap, and Hydra pre-installed, and prebuilt attack playbooks provide structured workflows for common assessments. The primary audience is bug bounty hunters, red teamers, and penetration testers who want an LLM-driven agent to coordinate real security tooling against live targets.

Lo que la gente pregunta sobre pentestagent