Domain Search MCP Server

Domain Search MCP

Naming engine with availability intelligence — an MCP server that scores the names your model generates and runs availability checks against domains, socials, and package registries. Works with zero configuration using public RDAP/WHOIS, and optionally enriches results with registrar pricing via a backend you control.

🆕 v1.12.0: name_project — a two-phase naming engine. Call it once to get generation instructions for your model, call it again with candidates[] to get anti-slop scoring, ranking, and live availability checks across domains, socials, and npm. See name_project below.

🆕 v1.10.0: GoDaddy public endpoint integration! Enhanced fallback chain (RDAP → GoDaddy → WHOIS) with premium/auction domain detection. Circuit breaker pattern ensures resilience.

🤖 v1.9.0+: AI-powered domain suggestions work out of the box! No API keys needed - suggest_domains_smart uses our public fine-tuned Qwen 7B-DPO model. Plus: Redis distributed caching and /metrics endpoint for observability.

Built on the Model Context Protocol for Claude, Codex, VS Code, Cursor, Cline, and other MCP-compatible clients.

Features

What It Does

• Check a single name across multiple TLDs.
• Bulk-check up to 100 names for one TLD.
• Compare registrar pricing (uses backend when configured).
• Suggest names and validate social handles.
• Detect premium/auction signals for search_domain.

How It Works

Availability and pricing are intentionally separated:

Availability Chain (zero-config):
┌─────────┐     ┌─────────┐     ┌─────────┐
│  RDAP   │ ──► │ GoDaddy │ ──► │  WHOIS  │
│ (fast)  │     │(premium)│     │(fallback│
└─────────┘     └─────────┘     └─────────┘

• Availability (default, no keys needed):
  • RDAP: Primary source - fast, unlimited, public registry data
  • GoDaddy: Secondary - adds premium/auction detection (30 req/min, circuit breaker protected)
  • WHOIS: Last resort fallback for edge cases
• Pricing (optional):
  • Recommended: PRICING_API_BASE_URL (backend with Porkbun keys)
  • Optional BYOK: Porkbun/Namecheap only when backend is not configured

This keeps the server zero-config while letting power users enable pricing.

Pricing Verification

Responses include price_check_url (registrar checkout/search link) and may include price_note when a price is estimated. Always verify the final price on the registrar checkout page before purchase.

If an auction/premium signal is detected, results include an aftermarket block with links to marketplace pages when available. Taken domains may include Sedo auction hints (public feed) and nameserver-based marketplace hints (Sedo/Dan/Afternic).

Quick Start

Option 1: npx (Recommended)

No installation needed - run directly:

npx -y domain-search-mcp@latest

Option 2: From Source

git clone https://github.com/dorukardahan/domain-search-mcp.git
cd domain-search-mcp
npm install
npm run build
npm start

Transport Options

stdio (Default)

For MCP clients like Claude Desktop, Cursor, VS Code - uses stdin/stdout:

npx -y domain-search-mcp@latest

HTTP/SSE (ChatGPT, Web Clients, LM Studio)

For ChatGPT Actions, web apps, and REST API clients:

# Start HTTP server on port 3000
npx -y domain-search-mcp@latest --http

# Or with custom port
MCP_PORT=8080 npx -y domain-search-mcp@latest --http

Endpoints:

• /mcp - MCP protocol (POST for messages, GET for SSE stream)
• /api/tools/* - REST API for each tool (ChatGPT Actions compatible)
• /openapi.json - OpenAPI 3.1 specification
• /health - Health check
• /metrics - Prometheus-compatible metrics (cache stats, request counts, AI inference health)

ChatGPT Custom GPT Integration

1. Start the HTTP server (see above)
2. Expose via ngrok: ngrok http 3000
3. In ChatGPT, create a Custom GPT and add an Action
4. Import the OpenAPI spec from https://your-ngrok-url.ngrok-free.dev/openapi.json
5. Test the tools!

For production deployment, use a permanent domain with SSL instead of ngrok.

REST API Example:

curl -X POST https://your-domain/api/tools/search_domain \
  -H "Content-Type: application/json" \
  -d '{"domain_name":"vibecoding"}'

MCP Client Config

Claude Code (.mcp.json in project root):

{
  "mcpServers": {
    "domain-search": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "domain-search-mcp@latest"]
    }
  }
}

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "domain-search": {
      "command": "npx",
      "args": ["-y", "domain-search-mcp@latest"]
    }
  }
}

💡 Tip: Always use @latest to ensure you're running the newest version with all features.

Tools

All 12 tools listed below are exposed to MCP clients by default. The 6-tool slim profile (name_project, search_domain, bulk_search, check_socials, tld_info, ai_health) is opt-in — set SLIM_TOOLS=true if you want a sharper tool-selection surface for simpler client integrations (see Environment Variables). A future 2.0 release may flip the default to slim.

ADVANCED_TOOLS=true is a deprecated alias that forces the full surface and overrides SLIM_TOOLS; it's a harmless no-op today since full is already the default.

name_project

Flagship two-phase naming engine. Call it once to get lane-by-lane generation instructions for your model; call it again with candidates[] to get anti-slop scoring, ranking, and live availability checks across domains, socials, and npm.

• Modes: brief (describe what you're naming), auto (analyze the current workspace), from_name (find domains/variants for a name you already like), from_domain (fit a project/brand to a domain you found).
• Phase 1 (no candidates): returns generation instructions + lane prompts.
• Phase 2 (candidates present): scores + ranks candidates, then checks availability for the top 12 against targets.tlds / targets.platforms — omit targets for pure naming with no availability calls.

Scores are heuristic rankings for comparing candidates against each other — not objective, universal brandability truth. Availability results reflect a single source checked at one moment in time; re-verify before you register or rely on anything.

Phase 1 — call with no candidates:

{"mode": "brief", "brief": "an MCP naming engine"}

Brief: an MCP naming engine

Now generate between 30 and 50 name candidates spread across these lanes:
- [evocative] Real words borrowed for their feeling, not their meaning (like Slack, Notion, Bolt). Single dictionary words preferred.
- [invented] Coined words that do not exist but sound like they could (like Zapier, Klarna). Must be pronounceable on first read.
- [compound] Two short real words fused (like Facebook, Snapchat). Both halves must stay readable; no glue letters.
- [premium] Short, expensive-feeling names: 4-7 letters, strong single or double syllable (like Stripe, Vercel, Arc).

Rules: single words or tight compounds, no taglines, no explanations yet. Then call name_project again with the SAME arguments plus candidates:[...] to get scoring and availability.

Phase 2 — resubmit the same arguments plus candidates:

{"mode": "brief", "brief": "an MCP naming engine", "candidates": ["Nexify", "Corda"]}

| Name | Score | Verdict | Badges | Why |
| --- | --- | --- | --- | --- |
| Corda | 97 strong | - | - | no AI-slop patterns; clean pronunciation and typing |
| Nexify | 60 middling | - | - | slop: overused prefix "nex-"; slop: overused suffix "-ify" |
2 candidates received, 2 passed constraints, top 2 returned.
No availability-check targets - pure naming mode.

Badges: tld✓ free to register, tld$ for sale (aftermarket/premium - registered or priced, not free to register), tld✗ taken, tld? unknown. ccTLD checks (.ai / .io / .sh / .ac) are cross-checked against native WHOIS/DNS ground truth, not taken on RDAP's word alone.

See docs/API.md for the full parameter/response schema.

Core Search

• search_domain: Check a name across multiple TLDs, adds premium/auction signals.
• bulk_search: Check up to 100 names for a single TLD.
• compare_registrars: Compare pricing across registrars (backend when configured).

AI-Powered Suggestions

• suggest_domains: Generate variations (prefix/suffix/hyphen).
• suggest_domains_smart: 🤖 AI-powered brandable name generation using fine-tuned Qwen 7B-DPO. Zero-config - works instantly!
• analyze_project: Scan local project or GitHub repo to extract context and suggest matching domain names.

Domain Investment

• hunt_domains: Find valuable domains for investment - scans Sedo auctions, generates patterns, calculates investment scores.
• expiring_domains: Monitor domains approaching expiration (requires federated negative cache).

Utilities

• tld_info: TLD metadata and restrictions.
• check_socials: Username availability across platforms.
• ai_health: Check status of AI inference services (VPS Qwen, circuit breakers, adaptive concurrency).

Configuration

Pricing Backend (Recommended)

Set a backend URL that owns registrar keys (Porkbun). The MCP will call /api/quote and /api/compare on that backend for pricing.

PRICING_API_BASE_URL=https://your-backend.example.com
PRICING_API_TOKEN=optional_bearer_token

Optional BYOK (Local)

Used only if PRICING_API_BASE_URL is not set.

• Porkbun keys:
  • https://porkbun.com/account/api
  • https://porkbun.com/api/json/v3/documentation
• Namecheap keys (IP whitelist required):
  • https://ap.www.namecheap.com/settings/tools/apiaccess/
  • https://www.namecheap.com/support/api/intro/

PORKBUN_API_KEY=pk1_your_api_key
PORKBUN_API_SECRET=sk1_your_secret
NAMECHEAP_API_KEY=your_api_key
NAMECHEAP_API_USER=your_username
NAMECHEAP_CLIENT_IP=your_whitelisted_ip

Redis Distributed Cache (Optional)

For horizontal scaling across multiple MCP instances, configure Redis:

REDIS_URL=redis://:password@host:6379

Without Redis, the server uses in-memory caching (works fine for single instances). Redis enables:

• Shared cache across multiple server instances
• Persistent cache surviving restarts
• Better cache hit rates in load-balanced deployments

AI Inference (bring your own endpoint)

AI-powered suggestions (suggest_domains_smart) use your own inference endpoint when configured. Point QWEN_INFERENCE_ENDPOINT at a llama.cpp/Qwen server you control. If it is unset, suggestions fall back to the built-in offline semantic engine (no external calls, no API keys needed).

# Public hosts must use HTTPS; loopback/private hosts may use HTTP.
QWEN_INFERENCE_ENDPOINT=http://127.0.0.1:8070
QWEN_API_KEY=optional_if_secured

Environment Variables

Output Format

Tool responses are returned as Markdown tables by default. If you need raw JSON for programmatic use, set:

OUTPUT_FORMAT=json

Data Sources

Pricing Behavior

• Live price is attempted first for every available domain.
• If live quotes fail or are rate-limited, the result falls back to the catalog estimate and includes price_note.
• Always verify pricing via price_check_url before purchase.

Examples

Basic Search (No API Keys)

search_domain("myproject", ["com", "io", "dev"])

┌─────────────────┬───────────┬─────────┬────────┐
│ Domain          │ Available │ Premium │ Source │
├─────────────────┼───────────┼─────────┼────────┤
│ myproject.com   │ ✅        │ No      │ rdap   │
│ myproject.io    │ ❌        │ -       │ rdap   │
│ myproject.dev   │ ✅        │ Yes     │ godaddy│
└─────────────────┴───────────┴─────────┴────────┘

AI-Powered Suggestions

suggest_domains_smart("coffee shop in seattle", { style: "brandable" })

→ seattlebrew.com, pugetperk.io, raincitycoffee.co, cascadiacafe.com

Bulk Check

bulk_search(["startup", "launch", "begin", "init"], "io")

→ Checks startup.io, launch.io, begin.io, init.io in parallel

Development

npm run dev       # watch mode
npm test          # run Jest
npm run build     # compile to dist/

Release

See docs/RELEASE.md for the canary -> latest publish flow. Tags like v1.2.24 trigger GitHub Releases + npm publish via CI.

Changelog

See CHANGELOG.md for release history.

Security Notes

• Do not commit API keys or .mcpregistry_* files.
• Without PRICING_API_BASE_URL (or BYOK keys), pricing is not available (availability still works).

Upgrading

For npx Users

If you use npx domain-search-mcp (without @latest), npx may cache an old version.

Fix: Update your MCP config to use @latest:

"args": ["-y", "domain-search-mcp@latest"]

Or clear the npx cache manually:

npx clear-npx-cache  # then restart your MCP client

For Source/Git Users

cd domain-search-mcp
git pull origin main
npm install
npm run build

Staying Updated

• Watch the repo: Click "Watch" → "Releases only" on GitHub to get notified of new versions.
• Check releases: See GitHub Releases for changelog and upgrade notes.
• npm page: npmjs.com/package/domain-search-mcp shows the latest version.

Architecture

For detailed system architecture diagrams, see docs/ARCHITECTURE.md:

• Transport layer (stdio vs HTTP/SSE)
• Tool execution flow
• Data source waterfall (RDAP → Pricing API → WHOIS)
• VPS deployment architecture
• AI suggestion flow
• MCP session lifecycle

Why This Tool?

FAQ

Q: Does this work without any API keys? A: Yes! Availability checking uses public RDAP and GoDaddy endpoints. Only pricing requires API keys.

Q: Which MCP clients are supported? A: Claude Desktop, Claude Code, VS Code, Cursor, Cline (stdio), and ChatGPT, LM Studio (HTTP/SSE).

Q: How accurate is premium domain detection? A: GoDaddy's public endpoint detects most premium and auction domains. Always verify on registrar checkout.

Q: Can I self-host the AI suggestions? A: Yes! Set QWEN_INFERENCE_ENDPOINT to your llama.cpp server running the fine-tuned model.

Links

• npm: npmjs.com/package/domain-search-mcp
• MCP Registry: registry.modelcontextprotocol.io
• Glama: glama.ai/mcp/servers/@dorukardahan/domain-search-mcp
• Context7: context7.com/dorukardahan/domain-search-mcp

Documentation

• Architecture - System design and data flow
• API Reference - Tool schemas and responses
• Configuration - Environment variables
• Workflows - Common usage patterns