Hosted products
Menu
Docs Tools Blog GitHub
Documentation menu

Get started

Use

Reference

Version 2.38.0

Agent Skill and Claude Code plugin

OpenOSINT ships as an Agent Skill and as a Claude Code plugin. An agent that has it can investigate a domain, IP address or organization with real, passive tools, and must cite the tool call behind every finding. It needs uv and network access. It needs no API key and no OpenOSINT AI provider: your agent is the model.

Install

Claude Code (plugin)

The plugin bundles both skills and a passive MCP server, pinned to release 2.38.0.

/plugin marketplace add OpenOSINT/OpenOSINT
/plugin install openosint@openosint
/openosint:osint example.com

Any agent that supports Agent Skills

Codex, Gemini CLI, Cursor and other Agent Skills clients can install the skills only (no MCP server):

npx skills add OpenOSINT/OpenOSINT

Then ask your agent in plain words, or invoke the skill by name, for example /osint example.com in Claude Code. See the namespacing note below.

By hand

Copy skills/osint/ and skills/exposure/ from the repository into the skills folder your agent reads (for Claude Code, ~/.claude/skills/ or .claude/skills/ in a project).

What the skills do

Skill Use it for What it runs
osint Recon on a domain, IP address or organization exposure, dns, rdap, whois, dorks, gdelt and ip as the target needs, then a Markdown report
exposure "How exposed is this domain?" openosint exposure DOMAIN: an A to F grade for email authentication, TLS, registration and security headers

Both call the CLI with --json through uvx --from openosint==2.38.0 openosint ..., so they work on any agent without an MCP server. When the openosint MCP server is already connected, the osint skill uses its tools instead; they return the same data.

The report has four parts: target and scope, findings (each with its citation), what was not checked or needs a key, and leads to verify.

The passive rules

The skills hold the agent to these rules:

  1. Passive by default. No --allow-active, no OPENOSINT_ALLOW_ACTIVE, and no scrape, username, email or subdomain brute-force unless you ask for that tool by name. If an active tool would help, the agent says so and asks.
  2. Authorization. Only targets you own or are authorized to assess. If the target looks like a private individual, the agent stops and asks what the legitimate purpose is. OpenOSINT is not designed for people search.
  3. Every finding cites its tool call, written as claim [tool: args]. No citation, no finding.
  4. Summaries are leads, not proof. The agent reports what a tool returned, and shows errors and empty results as such.
  5. A missing key is named, not skipped. The agent tells you which variable to set. See Configuration.

One thing still contacts the target: the exposure TLS and header checks make a connection on port 443. For zero contact, ask for exposure DOMAIN --checks email,registration (DNS and registry only, not graded). More in Safety.

Troubleshooting

uvx is not found

Install uv (curl -LsSf https://astral.sh/uv/install.sh | sh on macOS and Linux), then open a new terminal so it is on your PATH. Desktop apps often start with a minimal PATH: for the plugin's MCP server, or any MCP client, use the absolute path from which uvx (macOS, Linux) or where uvx (Windows) as the command.

/osint or /openosint:osint does nothing

Skills installed by the Claude Code plugin are namespaced by the plugin name, so they are /openosint:osint and /openosint:exposure. Skills installed on their own (npx skills add, or copied by hand) are not namespaced, so the same skill is /osint. If a command is not found, type / and look at the list to see which form you have.

The first run is slow, or fails offline

uvx downloads the pinned OpenOSINT release the first time and caches it. That needs network access; later runs start from the cache.

A tool says a key is missing

That is expected for the keyed tools (Shodan, Censys, VirusTotal, AbuseIPDB, HaveIBeenPwned). The default flow does not use them. See Configuration for the variable names, or Troubleshooting.

I want the noisy tools

Ask for the tool by name and set OPENOSINT_ALLOW_ACTIVE=1 yourself. Read the noise table first: those tools send traffic from your IP or to the target.

Edit this page on GitHub