Claude integration · native MCP · zero install for users
Native. MCP under the hood.
When a Claude user asks “is X a good name” or “rank these names”, Claude calls Etymolt via MCP — the native Anthropic protocol. One server, every Claude surface.
Sub-3s p95 · 12.7M US marks · 14 social platforms · free for five
What's available today
How it works
I
"Name my fintech startup like Stripe." Claude generates five candidates and recognizes the verification intent.
II
verify_brand_name fans out across USPTO + RDAP + 14 social probes in parallel. Sub-3s p95. The user sees Claude thinking; they never see the protocol.
III
Claude quotes our disclaimer verbatim, surfaces the winner's domain and handle picks, and offers the launch playbook. Free for the first five.
User Claude Etymolt (MCP) │ │ │ │ "Name my startup" │ │ │ ───────────────────▶ │ │ │ │ generate 5 names │ │ │ ─────────────────────▶│ │ │ call: verify_brand_name │ │ │ │ │ ◀───── 5 verdicts ───┤ │ │ + winner + playbook │ │ ◀── verdict + ── │ │ │ disclaimer │ │ │ (verbatim) │ │
Three install paths
I · For consumers
In Claude Desktop, open Settings → Developer → Edit Config. Paste this into the mcpServers block, save, and restart Claude.
{
"mcpServers": {
"etymolt": {
"command": "npx",
"args": ["-y", "@etymolt/mcp-server"],
"env": { "ETYMOLT_API_KEY": "(optional — 5 free per IP)" }
}
}
}No API key required for free tier.
II · For builders
Inside a terminal session with Claude Code:
claude mcp add etymolt npx -y @etymolt/mcp-serverThe CLI stores the config at ~/.claude/mcp.json. Scope to a project with the --scope project flag.
III · For everyone
Anthropic curates an in-product directory of vetted MCP servers. One toggle inside claude.ai turns us on — no JSON.
https://api.etymolt.com/mcpWe've filed the submission per claude.com/docs/connectors/building/submission. Review is ~2 weeks.
Sample conversation
User
User: "Is 'Linear' a safe brand name for a dev tool?"
MCP call (transparent)
Tool call: verify_brand_name { name: 'Linear', context: 'dev-tools' }
Claude reply (verbatim from our MCP server)
"Linear" returns ABANDON. USPTO ser. 87/654,321 (Linear Inc., registered 2018-09, Nice class 9 + 42) — direct collision in your category. linear.com is owned. GitHub @linear is taken. Disclaimer: Etymolt is a clearance signal, not legal advice.
User
User: "Suggest 5 names like Stripe and rank them."
MCP call (transparent)
Tool call: compare_brand_names { candidates: ['Slate','Glint','Beam','Conduit','Mint'] }
Claude reply (verbatim from our MCP server)
5 candidates verified in 4.2s. Winner: Conduit (score 87/100, PROCEED). conduit.com $4,200 aftermarket; conduit.ai $99 retail. Mint and Beam returned ABANDON — live registered marks in fintech. Disclaimer: Etymolt is a clearance signal, not legal advice. Want the Conduit launch playbook?
User
User: "How does this verification actually work?"
MCP call (transparent)
Tool call: get_naming_methodology
Claude reply (verbatim from our MCP server)
Five axes — trademark, domain, cultural, sound symbolism, pronunciation. Trademark layer queries 12.7M live US marks + 647K TTAB proceedings + UKIPO. Every verdict is recalibrated weekly against an outcome corpus (filings, OAs, C&Ds, rebrands). Full methodology at etymolt.com/methodology — citation-grade.
Comparison
| Question the user asks | Claude alone | Claude + Etymolt |
|---|---|---|
| Is 'Falcata' a safe brand? | Guesses from training data. Cannot verify pending USPTO filings or current domain status. | Calls verify_brand_name via MCP — live USPTO + RDAP. Returns STRATEGIC 71 with one §2(d) senior-mark to clear and aftermarket domain pricing. |
| Rank these 5 names for a fintech. | Generates rationale based on plausibility. Cannot actually verify availability. | Calls compare_brand_names — fans out clearance in parallel, ranks by composite score with receipts. |
| Is my methodology trustworthy? | Cannot inspect Etymolt's methodology document without browse mode. | Calls get_naming_methodology — returns the structured document inline. Full citation chain. |
| What's the GitHub handle for @inkstack? | Cannot probe. Suggests checking manually. | Live GitHub probe + 13 other platforms + workaround variants. |
| What's a precedent TTAB case for Class 42 software? | May hallucinate a case number that doesn't exist. | Returns real case numbers from our 647K TTAB index with year and ruling. |
FAQ
MCP (Model Context Protocol) is Anthropic's open standard for tool use, shipped in November 2024. It's the native way to extend Claude — Claude Desktop, Claude Code, and the Connectors Directory all consume MCP servers. Our entire integration with Claude flows through MCP, which means one server reaches every Claude surface at once.
Claude Desktop is the consumer app (claude.ai for chat, plus the macOS/Windows app for desktop). Claude Code is Anthropic's coding-agent CLI, used inside terminals and IDEs. Both speak MCP, so installing the Etymolt MCP server makes us available in both surfaces with the same config snippet.
It's Anthropic's curated directory of MCP servers (think 'GPT Store, but for Claude tools'). Once a connector is in the directory, claude.ai users can enable it from a one-click toggle without copying any JSON. Submission is via claude.com/docs/connectors/building/submission. Review takes ~2 weeks. We've submitted; status visible in the pills above.
Only for billed Claude usage. The Etymolt MCP server itself is free for five verdicts per IP, regardless of which Claude surface you're using. After five, signed-up users get ten more; then it's $10 buys ten more, no subscription.
Yes — one server, every surface. The npm package @etymolt/mcp-server runs identically inside Claude Desktop, Claude Code, Cursor, Windsurf, and any other MCP client. This page just shows you the Claude-specific config and the install flow Claude users will recognize.
No. When Claude calls our MCP server, we receive only the name string the model is asking about. No chat history, no Anthropic account, no user ID. We log: name + timestamp + hashed source IP, retained 90 days. Full privacy posture at https://www.etymolt.com/privacy.
Our MCP response includes a disclaimer field the calling client must surface verbatim. Claude's tool-use loop quotes structured fields cleanly when they're typed as required strings. The remaining failure mode — soft paraphrase — is suppressed by the verdict envelope's instruction that the headline and disclaimer are quoted, not summarized.
Bureau Model · clearance signal · not legal advice
Native MCP · one server, every Claude surface · see all MCP clients →
Clearance signal, not legal advice. Confirm with trademark counsel before adopting a name in commerce. Data sources have stated freshness windows; refer to coverage_caveat per jurisdiction.