How to Search for Domains With an AI Agent (Cross-Registrar Workflow)

Domain availability differs by registrar — not by which one "owns" the domain (a name is either taken globally or not), but by which TLDs each registrar carries and what they charge. The minimum useful set: Cloudflare (at-cost, a curated TLD subset through the Registrar API beta on its hosted MCP), Name.com (broad inventory, official server, sandbox by default), Porkbun (at-cost, broad TLDs, community server). If the team deploys on Vercel, its MCP is a fourth headless source with per-TLD pricing. GoDaddy and Namecheap add nothing outside ChatGPT.

For Claude Desktop, add all three to claude_desktop_config.json. Cloudflare's server is hosted at mcp.cloudflare.com/mcp: add it as a custom connector under Settings → Connectors, or bridge it with mcp-remote as below and sign in with OAuth when the browser opens. Do not use the npm package @cloudflare/mcp-server-cloudflare — it is an older Workers-only server with no Registrar access. Name.com takes a username + token pair (and hits its test environment until NAME_API_URL points at production), Porkbun an API key + secret. Keep credentials in env blocks, never in args.

{
  "mcpServers": {
    "cloudflare": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.cloudflare.com/mcp"]
    },
    "name-com": {
      "command": "npx",
      "args": ["-y", "namecom-mcp@latest"],
      "env": {
        "NAME_USERNAME": "your_user",
        "NAME_TOKEN": "your_token",
        "NAME_API_URL": "https://mcp.name.com"
      }
    },
    "porkbun": {
      "command": "uvx",
      "args": ["porkbun-mcp"],
      "env": {
        "PORKBUN_API_KEY": "pk1_...",
        "PORKBUN_SECRET_KEY": "sk1_..."
      }
    }
  }
}

Most domain searches start with a vague idea, not a known name. Ask the agent to brainstorm a list of candidates before any availability check — that turns a single registrar call into a structured workflow with five or ten parallel checks.

Brainstorm 10 domain name candidates for a new project
called "an AI agent that reviews pull requests". Mix
short, brandable names (.ai, .dev) with descriptive ones
(.so, .build). Return the list, then ask me to confirm
which ones to check.

The agent should call each registrar's availability operation for every candidate — Cloudflare's domain-check endpoint through its execute tool, Name.com's ManageDomains with operation check, Porkbun's domains_check_bulk — then compose the results into a single table. The killer feature is the price spread: the same .com name typically costs about the same everywhere, but a .dev or .ai can vary by 30% or more.

For each candidate from step 3, use the Cloudflare, Name.com,
and Porkbun MCPs to check availability. Return a table with:
domain, available (yes/no), and price at each registrar
(USD, first year). Sort by price ascending.

Once the comparison table is in the conversation, pick the domain you want and the registrar to buy it from (usually the cheapest one that carries the TLD). Switch back to a single-registrar flow for the actual registration — the buy step is covered in the dedicated tutorial linked below.

I want to register codereview.dev from the cheapest source
in the table. Before calling register, restate the exact
domain, the cost, and which registrar you will use. Wait
for me to reply "yes" before proceeding.