█████╗  ██████╗ ███████╗███╗   ██╗████████╗███████╗ ██████╗  █████╗ ███╗   ██╗
██╔══██╗██╔════╝██╔════╝████╗  ██║╚══██╔══╝██╔════╝██╔════╝██╔══██╗████╗  ██║
███████║██║  ███╗█████╗  ██╔██╗ ██║   ██║   ███████╗██║     ███████║██╔██╗ ██║
██╔══██║██║   ██║██╔══╝  ██║╚██╗██║   ██║   ╚════██║██║     ██╔══██║██║╚██╗██║
██║  ██║╚██████╔╝███████╗██║ ╚████║   ██║   ███████║╚██████╗██║  ██║██║ ╚████║
╚═╝  ╚═╝ ╚═════╝ ╚══════╝╚═╝  ╚═══╝   ╚═╝   ╚══════╝ ╚═════╝ ╚═╝  ╚═╝╚═╝  ╚═══╝
      made by JohnDoeAnon in collaboration with AnonymousITALIA

AgentScan

Read-only scanner for exposed AI / LLM / agent services.

AgentScan finds AI services listening on standard ports that answer without authentication, identifies what they are, and enumerates their API endpoints — so you can see exactly what is exposed. It never writes, never sends prompts, and never executes anything on the target: only HTTP GET requests are issued.

Use it only against assets you own or are explicitly authorised to test.

Features

• 31 service signatures — Ollama, vLLM, LocalAI, LiteLLM, llama.cpp, TGI, KoboldCpp, LM Studio, TabbyAPI, Aphrodite, Xinference, Text-Generation-WebUI, Flowise, Langflow, n8n, Dify, AnythingLLM, Open WebUI, Gradio, Streamlit, ComfyUI, MCP servers, Jupyter, LangServe, Haystack/hayhooks, OpenLLM, SuperAGI, Chainlit, Letta/MemGPT, AutoGen Studio.
• Endpoint enumeration from five sources: OpenAPI (/openapi.json), per-service catalog, the shared OpenAI-compatible surface, an aggressive path sweep, and fingerprinting evidence.
• Aggressive path sweep — 68 API prefixes × 152 segments (+ 27 well-known paths) to discover endpoints that are not in any catalog.
• Catch-all / SPA guard — detects servers that answer 200 to any path, so endpoint lists are never inflated and services are not misidentified.
• Response snippets — each live endpoint shows a one-line body preview.
• Reports — console, Markdown, JSON, and a CyberStrike-ready payload.
• Optional response dump — save the body of every 2xx response for manual inspection.
• Zero dependencies — standard library only, Python 3.8+.

Install

pip install agentscan-py

(run with the agentscan command)

Note: the distribution name is agentscan-py because agentscan is already taken on PyPI by an unrelated project.

Or from source:

git clone https://github.com/JohnDoeAnonITA/agentscan-py.git
cd agentscan-py
pip install .

Usage

# scan a single host on the built-in AI/agent ports
agentscan -t 10.0.0.5 --authorized

# a subnet
agentscan -t 10.0.0.0/24 --authorized --rate 100 --out-dir ./reports

# specific ports
agentscan -t 10.0.0.5 --authorized --ports 11434,3000,5678,8000

# targets from a file
agentscan --targets hosts.txt --authorized

--authorized is mandatory — it is your confirmation that you are allowed to test the supplied targets.

Output

--out-dir DIR writes agentscan-<timestamp>.json, .md, and .cyberstrike.json. You can also write a single format with --json, --md, or --cyberstrike.

By default only endpoints returning 2xx are displayed; pass --all-status to also show 401/403/405 (auth / method-not-allowed).

Key options

Run agentscan --help for the full list.

Target formats

10.0.0.5                 single IP (IPv4/IPv6)
box.lab.internal         hostname (DNS)
10.0.0.0/24              CIDR subnet (capped by --max-hosts)
10.0.0.1-10.0.0.50       IP range (explicit end)
10.0.0.1-50              IP range (last-octet shorthand)

Safety

• Read-only: only GET requests. POST-only routes are reported as HTTP 405 (proof the function exists) but never called.
• Rate-limited globally (--rate).
• Host cap on CIDR expansion (--max-hosts).
• Authorization gate (--authorized).

Output example

NO-AUTH! 127.0.0.1:11434  Ollama [high] 6/23 endpoints (2xx)
           GET  /api/tags      200  [{"name":"llama3:latest", ...}]
           GET  /api/version   200  "0.32.5"
           GET  /v1/models     200  {"object":"list","data":[ ... ]}

License

MIT