nameai-mcp

A remote MCP server exposing a handful of name.ai's public, no-auth-required APIs as tools: domain search/availability, WHOIS lookup, TLD registration pricing, and TLD requirements. See src/nameai_mcp/server.py for the tool definitions and src/nameai_mcp/nameai_client.py for the HTTP client that talks to name.ai.

Endpoints

• GET /health — health check.
• POST /mcp (and GET /mcp for the SSE stream) — the MCP streamable-http endpoint. This is the URL an MCP client/host connects to.

Configuration

• NAMEAI_API_BASE_URL — base URL of the name.ai API this server calls. Defaults to https://name.ai.
• HOST / PORT — bind address for the server itself. Default 0.0.0.0:8000.

Local development

uv sync
uv run uvicorn nameai_mcp.app:app --reload

Docker

docker build -t nameai-mcp .
docker run -p 8000:8000 nameai-mcp

The image runs nameai-mcp (the main() entry point in src/nameai_mcp/__init__.py), which reads HOST/PORT from the environment — pass -e PORT=... if your platform injects a different port.

Deploying

Any host that can run the Docker image works. To make it reachable by AI agents as https://mcp.name.ai, DNS for name.ai is on Cloudflare — add an A/CNAME record for the mcp subdomain pointing at wherever this container runs, and terminate TLS there (e.g. via Coolify/Traefik if hosted alongside the rest of nameaiv1).

Being reachable isn't the same as being discovered — nothing on name.ai currently points at this server. For an agent (or a scanner like ora) to find it, name.ai itself needs a pointer to it, e.g. a line in /llms.txt or a .well-known/mcp.json manifest. That's a change in the nameaiv1 repo, not this one.