AI Summary: PentestAgent is an AI-driven penetration testing tool designed to assist cybersecurity professionals in conducting thorough security assessments. It features multiple operational modes, including single-task assistance, autonomous task execution, and multi-agent orchestration, allowing users to adapt their approach based on project complexity. The tool supports integration with platforms like OpenAI and Anthropic, and can be run in Docker for enhanced isolation and access to a suite of pre-installed pentesting tools.

README

PentestAgent

AI Penetration Testing

https://github.com/user-attachments/assets/a67db2b5-672a-43df-b709-149c8eaee975

Requirements

• Python 3.10+
• API key for OpenAI, Anthropic, or other LiteLLM-supported provider

Install

# Clone
git clone https://github.com/GH05TCREW/pentestagent.git
cd pentestagent

# Setup (creates venv, installs deps)
.\scripts\setup.ps1   # Windows
./scripts/setup.sh    # Linux/macOS

# Or manual
python -m venv venv
.\venv\Scripts\Activate.ps1  # Windows
source venv/bin/activate     # Linux/macOS
pip install -e ".[all]"
playwright install chromium  # Required for browser tool

Configure

Create .env in the project root:

ANTHROPIC_API_KEY=sk-ant-...
PENTESTAGENT_MODEL=claude-sonnet-4-20250514

Or for OpenAI:

OPENAI_API_KEY=sk-...
PENTESTAGENT_MODEL=gpt-5

Any LiteLLM-supported model works.

Run

pentestagent                    # Launch TUI
pentestagent -t 192.168.1.1     # Launch with target
pentestagent --docker           # Run tools in Docker container

Docker

Run tools inside a Docker container for isolation and pre-installed pentesting tools.

Option 1: Pull pre-built image (fastest)

# Base image with nmap, netcat, curl
docker run -it --rm \
  -e ANTHROPIC_API_KEY=your-key \
  -e PENTESTAGENT_MODEL=claude-sonnet-4-20250514 \
  ghcr.io/gh05tcrew/pentestagent:latest

# Kali image with metasploit, sqlmap, hydra, etc.
docker run -it --rm \
  -e ANTHROPIC_API_KEY=your-key \
  ghcr.io/gh05tcrew/pentestagent:kali

Option 2: Build locally

# Build
docker compose build

# Run
docker compose run --rm pentestagent

# Or with Kali
docker compose --profile kali build
docker compose --profile kali run --rm pentestagent-kali

The container runs PentestAgent with access to Linux pentesting tools. The agent can use nmap, msfconsole, sqlmap, etc. directly via the terminal tool.

Requires Docker to be installed and running.

Modes

PentestAgent has three modes, accessible via commands in the TUI:

TUI Commands

/assist <task>    One single-shot instruction.
/agent <task>     Run autonomous agent on task
/crew <task>      Run multi-agent crew on task
/interact <task> Chat with the agent in guided mode
/target <host>    Set target
/tools            List available tools
/notes            Show saved notes
/report           Generate report from session
/memory           Show token/memory usage
/prompt           Show system prompt
/mcp <list/add>   Visualizes or adds a new MCP server.
/clear            Clear chat and history
/quit             Exit (also /exit, /q)
/help             Show help (also /h, /?)

Press Esc to stop a running agent. Ctrl+Q to quit.

Playbooks

PentestAgent includes prebuilt attack playbooks for black-box security testing. Playbooks define a structured approach to specific security assessments.

Run a playbook:

pentestagent run -t example.com --playbook thp3_web

Tools

PentestAgent includes built-in tools and supports MCP (Model Context Protocol) for extensibility.

Built-in tools: terminal, browser, notes, web_search (requires TAVILY_API_KEY), spawn_mcp_agent

Agent Self-Spawning (spawn_mcp_agent)

spawn_mcp_agent is a built-in tool that allows a running agent to spawn a child copy of itself as a subordinate MCP server connected over stdio. The child process is fully isolated — its own runtime, LLM client, conversation history, and notes store — and its complete tool set is injected back into the parent agent’s available tools after spawning.

This enables hierarchical, multi-agent workflows without any external orchestration: the agent self-organises by delegating scoped subtasks to children it spawns on demand.

After spawn_mcp_agent returns, the child’s tools (run_task, run_task_async, await_tasks, etc.) are available on the next tool call. The child’s server name is assigned automatically (e.g. child_agent_1) and returned in the result.

Example — orchestrator delegating parallel recon to two children:

# Turn 1: spawn two isolated child agents
spawn_mcp_agent  target="10.0.1.0/24"  scope=["10.0.1.0/24"]
spawn_mcp_agent  target="10.0.2.0/24"  scope=["10.0.2.0/24"]

# Turn 2: children's tools are now available — delegate work asynchronously
child_agent_1__run_task_async  task="Full port scan and service enumeration"
child_agent_2__run_task_async  task="Full port scan and service enumeration"

# Turn 3: wait and collect
child_agent_1__await_tasks  task_ids=["<id1>"]  timeout_seconds=600
child_agent_2__await_tasks  task_ids=["<id2>"]  timeout_seconds=600
child_agent_1__get_task_result  task_id="<id1>"
child_agent_2__get_task_result  task_id="<id2>"

MCP RAG Tool Optimizer

When an MCP server exposes more than 128 tools, PentestAgent automatically replaces the full catalogue with a single mcp_<server>_rag_optimizer tool. This meta-tool uses embedding similarity (via LiteLLM, default text-embedding-3-small) to retrieve the most relevant tools for the task at hand and injects them into the agent’s next turn — keeping the context window manageable without losing access to the full tool set.

The optimizer is transparent to the agent: it calls the RAG tool with focused natural-language queries describing what it needs, and the matching tools become available on the next turn to call directly.

Usage guidance for the agent:

Embeddings are computed once at startup and cached, so repeated queries are fast. The optimizer is built per-server, so each MCP server with a large catalogue gets its own independent index.

Tip: Pass one query per distinct capability rather than combining everything into one query. ["list open ports on a host", "get process memory usage"] retrieves better results than ["list ports and memory and CPU"].

MCP Integration

PentestAgent supports MCP (Model Context Protocol) in two directions: consuming external MCP servers as tool sources, and exposing itself as an MCP server so external clients (Claude Desktop, Cursor, etc.) can drive PentestAgent programmatically.

Consuming External MCP Servers (Client Mode)

Configure mcp_servers.json to connect PentestAgent to any external MCP servers. Example config:

{
  "mcpServers": {
    "nmap": {
      "command": "npx",
      "args": ["-y", "gc-nmap-mcp"],
      "env": {
        "NMAP_PATH": "/usr/bin/nmap"
      }
    }
  }
}

Exposing PentestAgent as an MCP Server (Server Mode)

PentestAgent can run as an MCP server, allowing any MCP-compatible client to submit tasks, inspect results, and control the agent remotely. Two transports are supported:

STDIO — for local clients (e.g. Claude Desktop, Cursor):

pentestagent mcp_server --type stdio
pentestagent mcp_server --type stdio --target 192.168.1.1 --scope 192.168.1.0/24
pentestagent mcp_server --type stdio --model claude-sonnet-4-20250514 --docker

SSE (HTTP) — for remote or networked clients:

pentestagent mcp_server --type sse
pentestagent mcp_server --type sse --host 0.0.0.0 --port 8080
pentestagent mcp_server --type sse --target 10.0.0.1 --scope 10.0.0.0/24 --docker

The SSE transport exposes a single /mcp endpoint supporting POST (requests), GET (persistent SSE stream for server-initiated push), and DELETE (session teardown). Sessions are tracked via the Mcp-Session-Id header.

All mcp_server flags:

Example: Claude Desktop config (claude_desktop_config.json)

{
  "mcpServers": {
    "pentestagent": {
      "command": "pentestagent",
      "args": ["mcp_server", "--type", "stdio"]
    }
  }
}

MCP Server Tools Reference

When acting as an MCP server, PentestAgent exposes the following tools:

Server Status & Config

Task Execution

Task Inspection

Task Control

Tool Management

Conversation History

Memory

Observability

Async Task Workflow Example

For long-running recon tasks, use the async pattern:

# 1. Submit tasks without blocking
run_task_async  task="Enumerate subdomains of example.com"  target="example.com"
run_task_async  task="Run nmap SYN scan on example.com"     target="example.com"

# 2. Block until both finish (up to 5 minutes)
await_tasks  task_ids=["<id1>", "<id2>"]  timeout_seconds=300

# 3. Retrieve full results
get_task_result  task_id="<id1>"
get_task_result  task_id="<id2>"

CLI Tool Management

pentestagent tools list         # List all tools
pentestagent tools info <name>  # Show tool details
pentestagent mcp list           # List MCP servers
pentestagent mcp add <name> <command> [args...]  # Add MCP server
pentestagent mcp test <name>    # Test MCP connection

Knowledge

• RAG: Place methodologies, CVEs, or wordlists in pentestagent/knowledge/sources/ for automatic context injection.
• Notes: Agents save findings to loot/notes.json with categories (credential, vulnerability, finding, artifact). Notes persist across sessions and are injected into agent context.
• Shadow Graph: In Crew mode, the orchestrator builds a knowledge graph from notes to derive strategic insights (e.g., “We have credentials for host X”).

Project Structure

pentestagent/
  agents/         # Agent implementations
  config/         # Settings and constants
  interface/      # TUI and CLI
  knowledge/      # RAG system and shadow graph
  llm/            # LiteLLM wrapper
  mcp/            # MCP client and server configs
  playbooks/      # Attack playbooks
  runtime/        # Execution environment
  tools/          # Built-in tools

Development

pip install -e ".[dev]"
pytest                       # Run tests
pytest --cov=pentestagent    # With coverage
black pentestagent           # Format
ruff check pentestagent      # Lint

Legal

Only use against systems you have explicit authorization to test. Unauthorized access is illegal.

License

MIT