The two-phase design is the interesting part: it does not generate names, your model does — this supplies the lane prompts, then scores what comes back and checks what survives. Premium and auction domains are flagged rather than reported as free, which is the failure mode that wastes the most time. Treat the scores as rankings for comparing candidates against each other, and the availability results as one source at one moment — worth re-checking before you register anything.
A naming engine with availability checking behind it. name_project hands your own model the generation instructions, scores what comes back for AI-slop patterns, and checks the survivors against domains, social platforms and npm. The other tools are that availability layer on its own — multi-TLD lookups, bulk checks, registrar price comparison, TLD metadata.
- A two-phase naming run: phase one returns lane-by-lane generation instructions for your model, phase two scores and ranks the candidates and checks availability for the top 12
- Four naming modes — from a brief, from the current workspace, from a name you already like, or fitted to a domain you found
- Anti-slop scoring that names its reason, such as an overused nex- prefix or -ify suffix
- One name checked across .com, .io, .dev, .ai and hundreds of other TLDs, with premium and auction domains flagged rather than reported as free
- Up to 100 names checked against a single TLD in one parallel call
- Registrar pricing compared for first year and renewal, each result carrying a price_check_url so the final number is confirmed at checkout
- Username availability across GitHub, npm, PyPI, Reddit, Twitter and other platforms, each with its confidence level stated
- Domain variations generated mechanically — hyphenated, numbered, prefixed — or as brandable names from a description
- A local project or GitHub repository scanned for context, then names suggested from what it found
- Aftermarket signals on taken domains: Sedo auction hints and nameserver-based marketplace hints
- TLD metadata: what it is for, its price range, its restrictions
- Investment scans: pattern-generated candidates scored, and domains approaching expiry
Npx -y domain-search-mcp@latest over stdio, or --http for a server on MCP_PORT 3000 that also exposes REST endpoints and an OpenAPI 3.1 spec for ChatGPT Actions. Availability needs no keys at all: RDAP first, GoDaddy for premium and auction detection, WHOIS as the last resort. Pricing is the part that needs setting up — point PRICING_API_BASE_URL at a backend that holds registrar keys, or bring your own with PORKBUN_API_KEY and PORKBUN_API_SECRET, or NAMECHEAP_API_KEY, NAMECHEAP_API_USER and NAMECHEAP_CLIENT_IP. SLIM_TOOLS=true cuts the surface to six tools; REDIS_URL shares the cache across instances; QWEN_INFERENCE_ENDPOINT points brandable suggestions at your own inference server, with an offline semantic engine as the fallback when it is unset.
One command — npx -y domain-search-mcp@latest