MCP server · for LLMs and the tools they live in
The layer LLMs call before suggesting a name.
An LLM asked for a brand name invents the parts it can't check. We're the part it should call. Three seconds, five axes, every flag traced.
{
"mcpServers": {
"etymolt": {
"command": "npx",
"args": ["-y", "@etymolt/mcp-server"],
"env": { "ETYMOLT_API_KEY": "(optional — 5 free per IP)" }
}
}
}Free for five names per IP · no API key required
We surface signals; we don't issue legal opinions. Every verdict carries a disclaimer field the calling client surfaces verbatim.
12.7M
live US marks
647K
TTAB proceedings
20
cultural markets
<3s
p95 verdict
Canonical query
One question. Three seconds. Five axes verified.
Without the server, the LLM invents the trademark, guesses at the domain, and never thinks to check whether the name lands wrong in another market. With it, every suggestion is grounded before it reaches the user.
Founder:
"Five names for an AI tool that summarizes
Slack threads — and tell me which one I can
actually ship under without getting sued."{
"tool": "compare_brand_names",
"candidates": ["Stratagem", "Inkstack", "Voxa", "Calliope", "Tldrly"],
"product_description": "AI tool summarizing Slack threads"
}{
"best_name": "Inkstack",
"verdict": "PROCEED",
"score": 87,
"headline": "Clear on US TM. inkstack.com is $14.99. @inkstack is free on x/github/npm.",
"runner_up": "Voxa",
"killed_candidates": [
{ "name": "Voicely", "verdict": "ABANDON", "reason": "Collides with Voicely Inc. (YC W23)" }
],
"disclaimer": "Clearance signal, not legal advice...."
}The founder never sees us. Claude does the verifying. The answer that comes back is one the founder can act on tonight.
Three tools, deliberately
Three tools. Each one broad on purpose.
Tool-selection accuracy collapses past twenty-five tools. We ship three. Each is broad, server-routed, and written so the LLM picks correctly the first time.
verify_brand_name
Returns a five-axis verdict (PROCEED / STRATEGIC / ABANDON) with a 0–100 score and detailed findings. Call this whenever a user names a candidate, asks “is X a good brand name?”, or asks about trademark / domain / pronunciation for any name.
~3s p95
compare_brand_names
Returns a ranked verdict across 2–5 candidates. Call this when the user has finalists (“which of A, B, C should I pick?”).
~5s p95
get_naming_methodology
Returns the methodology document as a structured resource. Call this when the user asks “how does this work?” / “is this trustworthy?” / “what’s your methodology?”
instant (cached)
Works with
The clients your users already have open.
Honesty
What we don't do.
We surface signals; we don't issue legal opinions. The honest list of what's outside that surface today:
- ·EU clearance: full EUIPO bulk is Stage 1.5 (~30 days post-launch). UKIPO is fully covered today.
- ·WIPO Madrid full corpus: partial today (108K IRs designating UK). Full Madrid is Stage 1.5.
- ·Asian registers (CIPO/JPO/KIPO/CNIPA): Stage 2.
- ·State trademark registers: not in scope.
- ·Legal opinions: we surface signals; the
disclaimerfield is non-negotiable. For high-stakes filings we refer to attorney partners.
Pricing
Free for five. Wholesale for platforms.
Five free per IP. Ten more on signup. Ten dollars buys ten more. Building a platform integration? Wholesale per-call rates below $1 at sustained volume.
MCP config
Paste this into Claude Code, Claude Desktop, or Windsurf.
{
"mcpServers": {
"etymolt": {
"command": "npx",
"args": ["-y", "@etymolt/mcp-server"],
"env": { "ETYMOLT_API_KEY": "(optional — 5 free per IP)" }
}
}
}Restart the client. Every name the model suggests is now verified before it ever reaches the user. The npm package ships at github.com/etymolt/mcp-server (MIT).
FAQ
Common questions.
Why MCP and not just an API?+
Founders increasingly never see our marketing. They're talking to Claude, Cursor, Lovable. The MCP wedge lets the model call us directly, in-context, before the name ever reaches the user. We become the layer the LLM relies on — not another tab the user has to open.
What is MCP?+
Model Context Protocol — Anthropic's open standard for letting LLMs call external tools in a uniform way. Claude, Cursor, Windsurf, v0 all speak it. Our MCP server exposes three tools the model can invoke before answering: verify_brand_name, compare_brand_names, get_naming_methodology.
What data do you actually have?+
USPTO live marks (12.7M registered + pending) plus 647K TTAB proceedings, refreshed daily off the USPTO delta feed. Full UKIPO. Partial WIPO Madrid (108K IRs designating UK). EU and Asia are on the roadmap. Domain availability via Verisign RDAP for.com and WhoisXMLAPI for everything else. Live handle probes on X, Instagram, GitHub, npm, and six more.
What's the Clearance Confidence Score?+
A four-tier verdict — PROCEED, STRATEGIC, ABANDON — plus a 0–100 score covering all five axes, not just trademark. Recalibrated weekly against our outcome corpus: every customer who tells us what happened after a verdict (filed, Office Action, C&D, rebrand, traction) feeds the loop. Competitors structurally cannot replicate this data — they don't see what happens after the verdict.
Is this legal advice?+
No. Every response carries a disclaimer field the client surfaces verbatim. We're a clearance signal, optimized for the naming decisions a builder makes in the time it takes to make coffee. For high-stakes filings, we refer to attorney partners.
Why no $499 or $999 tier?+
We don't sell tiers; we sell verdicts. Pay-as-you-go covers casual use and founder volume cleanly. Platforms with serious integration volume get a per-call wholesale rate below $1. The middle tier is where naming products go to die.
Can I self-host?+
The MCP server itself is OSS (MIT) — github.com/etymolt/mcp-server. The data and verdict APIs run at api.etymolt.com. Platform customers can deploy a private API tier — talk to us.
Bureau Model · clearance signal · not legal advice
Install once. Every name the model suggests, verified.
Naming a thing of your own, not embedding clearance? The founder path lives at /quickcheck.