Domain Finder

Search and recommend good domain names to buy. Describe a product or idea; the app brainstorms brandable names (AI + rule-based combos), checks real availability via RDAP, and returns a ranked list of buy candidates.

Source-available under the Elastic License 2.0 (SPDX-License-Identifier: Elastic-2.0) — see License.

Connect to the MCP server

Domain Finder is also a remote MCP server — check a brand name across domains and the GitHub / npm / PyPI namespaces from any MCP-capable agent.

• Endpoint: https://domain-finder-theta.vercel.app/api/mcp (Streamable HTTP)
• Tools:
  • check_name — check whether one or more candidate names are free across domains and the GitHub / npm / PyPI namespaces, in a single call.
  • generate_names — generate candidate names from a description, each pre-checked for domain availability; feed the favorites into check_name.
• Read-only, and no credentials required — nothing to sign up for, no keys to paste. Safe to connect.

Cursor (~/.cursor/mcp.json) and any client with native Streamable-HTTP support:

{
  "mcpServers": {
    "domain-finder": {
      "type": "streamable-http",
      "url": "https://domain-finder-theta.vercel.app/api/mcp"
    }
  }
}

Claude Desktop (claude_desktop_config.json) — bridge a remote server with mcp-remote:

{
  "mcpServers": {
    "domain-finder": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://domain-finder-theta.vercel.app/api/mcp"]
    }
  }
}

How it works

query ─▶ generate ─▶ check availability ─▶ rank ─▶ results
         (AI + rules)   (RDAP)              (0–100 score)

• Generation (lib/generate.ts) — merges three sources of candidates:

  • Rule-based combos: keyword extraction (stopword-filtered) × prefixes/ suffixes (get, try, -ly, -hq, -app, …), shorter-first.
  • AI brainstorm (optional): Vercel AI SDK via AI Gateway (generateObject
    • zod). Degrades gracefully to rule-based when no API key is present.
  • Domain hacks (lib/hacks.ts): read the word across the dot — delicio.us, bit.ly, recip.es. Built from IANA's full TLD list; needs no API. These ignore the TLD filter by design (the word picks the zone), so they have their own toggle.
  • Labels are deduped, capped, and expanded across the selected TLDs (round-robin so every label gets its .com before any gets a second TLD).

• Availability (lib/availability.ts) — RDAP behind a swappable AvailabilityProvider interface. Resolves each TLD to its authoritative registry RDAP server via the IANA bootstrap file (+ curated overrides), then queries the registry directly. Retries on 429.

• Status taxonomy (lib/rdap-status.ts) — "taken" isn't one thing. Modelled on Domainr's taxonomy, each domain gets a status and a coarse bucket:

  bucket	statuses	meaning
  registrable	available	buy it now, at retail
  dropping	deleting, expiring	in pendingDelete/redemption — backorder territory
  aftermarket	parked	owned, but nameservers say "for sale"
  unavailable	active, reserved	registered and in use
  unknown	unknown	no RDAP server for the TLD, or an error

  All derived free from RDAP's RFC 8056 status codes, expiration event, and nameservers. deleting domains also get an estimatedDropAt (pendingDelete is a fixed ~5-day window).

• Ranking (lib/rank.ts) — a 0–100 score from availability (45), TLD desirability (20), length (20), cleanliness/no-hyphens-or-digits (10), and a brandability nudge (5, hack > AI > combo). Sorted by bucket first, then score — so "buy it now" always outranks "might be gettable".

Swapping RDAP for a paid registrar API later (pricing + real buy flow) only touches lib/availability.ts — nothing else in the pipelin

... View full README on GitHub