Skip to content

Back to the catalog

Cybermes
github.comrepo

Cybermes

Cybermes is an open-source bug bounty and pentest automation framework that pairs a native Go MCP server (cybermes-mcp) with an autonomous Hermes CLI agent, orchestrating recon tools (subfinder, httpx, katana, ffuf, Nuclei, sqlmap), 200+ offensive playbooks covering IDOR/BOLA, JWT, SSRF, XSS, SQLi and cloud misconfig, and a validation-gate workflow that writes non-destructive PoC scripts before compiling findings into Markdown/HTML/PDF reports. It exposes its tools over JSON-RPC 2.0 to AI coding assistants (Claude Desktop, Cursor, Windsurf, Cline, Gemini/Antigravity) so an external LLM can drive scoped, evidence-backed security assessments.

TL;DR Go-based MCP server + autonomous CLI agent that automates recon, exploitation playbooks, and report generation for authorized bug bounty engagements.
Open original ↗
Document

What it is

Cybermes ships two distinct execution modes that differ in who does the reasoning and where code runs — a detail the summary flattens. The page also gives concrete install paths (npx, source, Docker) and a self-contained way to evaluate it without pointing it at a live target.

How it works

The two workflows have different trust properties:

  • Autonomous Hermes CLI — the local hermes.exe/cybermes engine is the reasoning brain. It maps the attack surface, spawns subprocesses (subfinder, httpx, katana, ffuf), filters their streamed output through smart_pipe, writes and runs PoC scripts, and executes validation loops directly on the OS. Raw tool output is archived to recon/; only relevant endpoints, status codes, and leaked secrets reach the model context.
  • MCP server — cybermes-mcp is a native Go JSON-RPC 2.0 stdio server exposing 10+ tools and 200+ SOPs. Here the external AI assistant (Cursor, Claude Desktop, Gemini/Antigravity, Windsurf, Cline, etc.) is the reasoning engine; Cybermes only executes tools, does local knowledge lookups, and compiles reports.
flowchart LR
    Op([Operator prompt]) --> Mode{Which mode?}
    Mode -->|Hermes CLI| H[Local engine reasons<br/>+ spawns subprocesses on OS]
    Mode -->|MCP| M[cybermes-mcp executes only;<br/>external IDE model reasons]
    H --> R[reports/ + PoCs]
    M --> R

Preconditions / Affected

Requires an .env with API keys, and the external recon binaries on PATH. python tools/doctor.py verifies system dependencies, path bindings, and tool integrity. MCP registration is a standard mcpServers block invoking npx -y cybermes-mcp.

What to do

Evaluate against the bundled mock app before touching any real scope:

python examples/mock_vulnerable_app.py   # http://127.0.0.1:8888
./cybermes "Assess http://127.0.0.1:8888"

Diagnose/repair the environment:

python tools/doctor.py
python tools/doctor.py --fix

Install (pick one): npx -y cybermes-mcp install (auto-detects clients; --kilo, --gemini, --cursor, --global, --dry-run flags), source ./setup.sh / .\setup_windows.ps1, or docker compose up -d.

Caveats

The tool is self-described as for authorized bug bounty and offensive use — the CLI mode executes generated PoC scripts and spawns tools directly on the host OS, so run it in an isolated environment. The page states PoC scripts are "non-destructive," but that is the author's claim, not a verified property; review generated pocs/poc_*.py before trusting them against real targets.

Technologies named here

Techniques named here

Related in Offensive Security