MCP server

The Model Context Protocol is how an AI assistant reaches tools outside itself. The Namesight MCP server is a small program that speaks it: the assistant asks for a name check, the server calls the same public API you would call yourself, and hands back the result.

It holds no data of its own. Scopes, credits and rate limits are enforced by the API exactly as they are for a direct request, so the server cannot do anything your key could not do.

You need an API key with the scans:read, scans:write and reference:read scopes. Create one under Settings → API keys.

Claude Desktop — add the server to claude_desktop_config.json:

{
  "mcpServers": {
    "namesight": {
      "command": "node",
      "args": ["/path/to/namesight/apps/mcp/dist/main.js"],
      "env": {
        "NAMESIGHT_API_KEY": "ns_live_...",
        "NAMESIGHT_API_URL": "https://api.namesight.app"
      }
    }
  }
}

Claude Code — one command:

claude mcp add namesight \
  --env NAMESIGHT_API_KEY=ns_live_... \
  --env NAMESIGHT_API_URL=https://api.namesight.app \
  -- node ./apps/mcp/dist/main.js

Only namesight_scan_name spends anything, and it is the only tool marked as writing, so a client that asks before write actions will ask before a scan.

The reference lists are also exposed as resources — namesight://reference/nice-classes, regions, tlds and adapters — so a client can attach them to the conversation once instead of spending a tool call each time. The check_brand_name prompt is a ready-made brief: scan the name, retry what could not be verified, explain the result, and propose alternatives when it is taken.

Every tool declares an output schema, and list_scans pages with cursor and status. A retry after a dropped connection is safe: the server replays the first answer instead of charging twice.

Scans take time, and a tool call returns once. By default namesight_scan_name returns the scanId immediately and the assistant reads the result with namesight_get_scan.

Pass waitMs to have the server wait instead: it polls until the scan finishes, starting about once a second and widening to five, then returns the result inline. If the budget runs out you get the partial scan rather than an error — the checks that did finish are still worth reading.

Once the server is connected you ask in plain language; the assistant picks the tools. A useful request names the classes or the markets, because that is what separates a real conflict from an unrelated mark.

Check whether "Lumina Labs" is free for a game studio
in Türkiye and the EU — Nice classes 9 and 41.

The assistant will typically look the classes up with namesight_reference, start the scan, wait for it, and summarise what blocks the name and what does not.

API failures come back as readable tool errors rather than crashes, so the assistant can explain the problem and, where it can, fix it and retry.

• 401 — the key is missing, unknown, revoked or expired
• 402 — the workspace plan does not include the API
• 403 — the key lacks a scope, or credit spending is switched off
• 429 — rate limit or credit balance exceeded