HexStrike AI setup: an MCP server, and the model lives in your agent
HexStrike AI does not pick the model. It is an MCP server that exposes security tools to an agent you already run, and that agent uses whatever model you configured.
Point Claude Code or OpenCode at your API, then add this server. The README calls it an "AI-Powered MCP Cybersecurity Automation Platform."
An operator-in-the-loop tool, not a second model
Use HexStrike when you drive the test and hand the agent tools over MCP. Strix and Shannon plan a scan on their own. Run it in a testing VM. (HexStrike AI) Set the model on the agent. Kimi K3.1 Pro is the pick for judgment.
Start the server, then register it with the agent
Install and MCP snippets from the HexStrike README, read November 18, 2026.
Clone and install the Python package
The README asks for Python 3 and the tools you intend to call. Install only what the scope needs.
git clone https://github.com/0x4m4/hexstrike-ai.git cd hexstrike-ai python3 -m venv hexstrike-env source hexstrike-env/bin/activate pip3 install -r requirements.txt python3 hexstrike_server.py --port 8888
HexStrike AI README. The documented health check is curl http://localhost:8888/health.
Add it to Claude Code as a stdio server
-- separates Claude's flags from the server command. Claude Code MCP docs.
claude mcp add --transport stdio hexstrike-ai -- python3 /absolute/path/hexstrike-ai/hexstrike_mcp.py --server http://localhost:8888Then set the model with the Claude Code setup.
Or add it to OpenCode
A local server is a command after --. OpenCode MCP docs.
opencode mcp add hexstrike -- python3 /absolute/path/hexstrike-ai/hexstrike_mcp.py --server http://localhost:8888Only on systems you own, a signed scope, a CTF or a lab. A custom model does not widen the scope.
Name the authorization and the limit
I am authorized to test only the lab host in SCOPE.md. From the nmap file already saved, rank the services a tester should review first. Do not touch a host outside SCOPE.md. Do not write exploit code.