# For AI agents

whoisbrowsing is built so a coding agent (Claude Code, Cursor, Codex and others) can add it to a project with no one signing up first. The agent creates a site, puts one script tag in the right file, and hands you two links: one to watch your visitors live, and one to keep the site under your Google sign-in.

## The prompt

Give your agent this prompt:

```
Add whoisbrowsing (live visitor view) to this project. Follow https://whoisbrowsing.com/SKILL.md exactly.
```

## SKILL.md

[SKILL.md](https://whoisbrowsing.com/SKILL.md) is one Markdown file with everything an agent needs: how to create the site, where the tag goes for each framework, a Content-Security-Policy note, how to verify, the two links for the human, one line for the project's AGENTS.md or CLAUDE.md, and the privacy promise. It is also an Agent Skill file, with name and description frontmatter.

## MCP server

`https://whoisbrowsing.com/mcp` is a remote MCP server over streamable HTTP, with no auth, listed in the official MCP Registry as `com.whoisbrowsing/mcp`. Its tools, in the order an agent uses them:

- `create_site({domains, name?})`: creates a site with no sign-up; the same answer as the API: snippet, watch links, claim link.
- `get_install_instructions({platform})`: where the tag goes on one platform (the install guide's Markdown), or the list of platforms.
- `check_live({siteId, key})`: whether the script is connected right now: websites, visitors and whether the owner is watching. key is the owner key after #wib= in the watch link.

### Claude Code

In your terminal ([Claude Code MCP docs](https://code.claude.com/docs/en/mcp)):

```bash
claude mcp add --transport http whoisbrowsing https://whoisbrowsing.com/mcp
```

### Cursor

In `.cursor/mcp.json` in the project, or `~/.cursor/mcp.json` for every project ([Cursor MCP docs](https://cursor.com/docs/context/mcp)):

```json
{
  "mcpServers": {
    "whoisbrowsing": {
      "url": "https://whoisbrowsing.com/mcp"
    }
  }
}
```

### Codex

In your terminal ([Codex MCP docs](https://developers.openai.com/codex/mcp)):

```bash
codex mcp add whoisbrowsing --url https://whoisbrowsing.com/mcp
```

Or in `~/.codex/config.toml`:

```toml
[mcp_servers.whoisbrowsing]
url = "https://whoisbrowsing.com/mcp"
```

## Agent API

`POST https://whoisbrowsing.com/api/agent/sites` creates a site with no sign-up and no auth:

```bash
curl -sS -X POST https://whoisbrowsing.com/api/agent/sites \
  -H 'Content-Type: application/json' \
  -d '{"domains":["localhost:3000","example.com"]}'
```

| Field | What it is |
|---|---|
| `siteId` | The site ID in the tag's `data-site`. Public |
| `snippet` | The script tag for every page |
| `watchUrls` | One watch link per website. Private: it holds an owner key |
| `claimUrl` | A one-time link to keep the site in your account. Expires after 30 days |
| `installDocs` | Where the tag goes on each platform |
| `next` | The remaining steps, in plain words |

Errors are JSON with a `hint`: `400` for a bad body, `429` past the limits (10 sites per network per hour, 5 unclaimed sites per browser). The spec: [openapi.json](https://whoisbrowsing.com/openapi.json), OpenAPI 3.1.

## The two links are for you, not the code

The watch link opens your own site with the live view on. It holds a private owner key, so treat it like a password. The claim link keeps the site in your account once you sign in with Google; it works once and expires after 30 days. A good agent hands you both in its reply and never writes them into the repository. Only the tag goes in the code.

## llms.txt and Markdown twins

- [llms.txt](https://whoisbrowsing.com/llms.txt): an index of every page and guide, as Markdown links.
- [llms-full.txt](https://whoisbrowsing.com/llms-full.txt): every guide in one file.
- Markdown twins: add `.md` to any article, install guide or main page for its Markdown source, for example https://whoisbrowsing.com/install/nextjs.md, https://whoisbrowsing.com/learn.md or https://whoisbrowsing.com/for-agents.md.

## Privacy, word for word

- Your visitors' browsers send the page, pointer position, clicks, scroll, device type, source, browser and operating system, and time on site. Never keystrokes, form values or page text.
- Nothing about your visitors is stored on our servers.
- No cookies on your visitors.
- Visitors from the EU/EEA, UK and Switzerland, and browsers with Global Privacy Control, are never shown.

More: [Install with an AI agent](https://whoisbrowsing.com/install/ai-agent.md), [How it works](https://whoisbrowsing.com/how-it-works.md), [Privacy](https://whoisbrowsing.com/privacy).