Guide3 min read

Claude Code + Ornold MCP: Full Setup Guide

Install Node.js, set up Claude Code, connect Ornold MCP, and automate antidetect browsers with AI. macOS and Windows.

Apr 17, 2026

One Prompt — Claude Code Installs Everything

You don't need to know what flags to use, which port your antidetect runs on, or what "vision mode" means. Paste one prompt with your token. Claude Code reads the public install guide on GitHub, asks you 2 questions in plain language, then runs the right install command for your setup.

How it works: Ornold publishes a public agent guide at github.com/ornold-ai/ornold-mcp. Claude Code fetches `AGENTS.md`, learns the install rules, and asks you only what it needs. No manual JSON editing, no remembering flags.

Step 1 — Get Your Token

Sign up at mcp.ornold.com
Open the Tokens page
Click Create token, copy the value (starts with `orn_sk_`)

Step 2 — Paste This Prompt into Claude Code

Open Claude Code (run `claude` in your terminal) and paste this whole block:

Install Ornold MCP for me. I'm new to this, so guide me.

1. Fetch and read the official install guide:
   https://raw.githubusercontent.com/ornold-ai/ornold-mcp/main/AGENTS.md

2. Then ask me these questions in plain language:
   a) Which antidetect browser do I use? Options: Linken Sphere, Dolphin Anty, Vision Browser, Wadex, or none yet (I just want to test).
   b) Which interaction mode do I want? Briefly explain the difference between "vision" (AI sees the page like a human, recommended for antidetect) and "dom" (AI reads HTML, faster but easier to detect). Recommend vision unless I have a reason to pick dom.
   c) If I picked Dolphin Anty or Vision Browser, ask me for the extra credentials they need (Dolphin API token / Vision token).

3. After I answer, install for Claude Code using the exact command from AGENTS.md, with my answers filled in. My Ornold token is: YOUR_TOKEN

4. Also install the AI skill: npx mcp-ornold-browser --install-skill

5. Verify the install by calling the browser_status tool. Tell me clearly whether it worked or not.

If anything fails, explain the error in plain language and suggest a fix.

Replace `YOUR_TOKEN` with your actual token before sending. Claude Code will fetch the guide, ask you 2-3 simple questions, run the install, and confirm it works.

What Claude Code Will Ask You

You don't need to memorize flags — the agent does. But here's a preview of the conversation so you're not surprised:

Question 1: Which antidetect browser?

**Linken Sphere** — desktop app for managing browser profiles. Local API on port 40080 (default).
**Dolphin Anty** — desktop app + cloud. Local port 3001 + your API token from the Dolphin dashboard.
**Vision Browser** — cloud service. You provide their token (starts with `vx_`).
**Wadex** — desktop app. Local API on port 8080.
**None / not sure** — install without antidetect for now. You can add one later.

Question 2: Vision or DOM mode?

**Vision (recommended)** — the AI takes a screenshot, identifies elements visually, then clicks by coordinates. Mimics human behavior, bypasses bot detection. Required for flow recording (record once, replay across many profiles).
**DOM** — the AI reads HTML and clicks via CSS selectors. Faster, cheaper, but anti-fraud systems detect this pattern. Use only when stealth doesn't matter (your own internal sites, scraping public data).
**Both** — install two separate MCP servers (one for vision, one for DOM). Pick this only if you have a clear reason — having both loaded confuses the agent (too many similar tools).

If you're unsure, pick **vision**. It's the safe default for antidetect work.

Step 3 — Verify

Claude Code will run `browser_status` automatically. To double-check, ask:

List my connected browsers using browser_list

If your antidetect browser is running, you'll see its profiles. You're done — start automating.

Already Know What You Want? Skip the Questions

If you already know your setup, paste the exact command. Examples:

Linken Sphere + vision (most common)

claude mcp add --transport stdio ornold-browser -- npx mcp-ornold-browser --token YOUR_TOKEN --linken-port 40080 --mode vision
npx mcp-ornold-browser --install-skill

Dolphin Anty + vision

claude mcp add --transport stdio ornold-browser -- npx mcp-ornold-browser --token YOUR_TOKEN --dolphin-port 3001 --dolphin-token YOUR_DOLPHIN_API_TOKEN --mode vision
npx mcp-ornold-browser --install-skill

Vision Browser (cloud) + vision mode

claude mcp add --transport stdio ornold-browser -- npx mcp-ornold-browser --token YOUR_TOKEN --vision-token YOUR_VISION_TOKEN --mode vision
npx mcp-ornold-browser --install-skill

Both modes (advanced)

Run vision and DOM as two separate servers so the agent doesn't mix tools:

claude mcp add --transport stdio ornold-vision -- npx mcp-ornold-browser --token YOUR_TOKEN --linken-port 40080 --mode vision
claude mcp add --transport stdio ornold-dom -- npx mcp-ornold-browser --token YOUR_TOKEN --linken-port 40080 --mode dom
npx mcp-ornold-browser --install-skill

Public Resources

**GitHub** — github.com/ornold-ai/ornold-mcp — the public agent install guide (`AGENTS.md`). Any AI agent can read it.
**npm** — mcp-ornold-browser — the package itself.
**Dashboard** — mcp.ornold.com — tokens, billing, team management.

Troubleshooting

**`claude mcp list` doesn't show ornold-browser** — re-run the install command. Make sure `YOUR_TOKEN` was replaced with the real token.
**Token invalid** — re-copy from mcp.ornold.com Tokens page (full string, including `orn_sk_` prefix).
**Cannot connect to browser** — your antidetect browser must be running, and the port flag must match its actual port.
**npx: command not found** — install Node.js 18+ from nodejs.org, then restart your terminal.
**Agent installed wrong flags** — paste the explicit command from "Skip the Questions" above. Or remove and reinstall: `claude mcp remove ornold-browser`.

What's Next?

Linken Sphere setup for advanced session management
Parallel browser automation — run 10+ sessions at once
CAPTCHA solving guide — reCAPTCHA, hCaptcha, Turnstile
Vision AI vs CSS selectors — why coordinate clicks beat DOM scraping

Explainer