Claude Code · agentic loop · native MCP
Use Etymolt in Claude Code.
npm init step. Verified name. Commit-safe.
When Claude Code runs npm init or scaffolds a startup, Etymolt fires before the project name gets committed — trademark, domain, 14 handles, cultural meaning, sound symbolism — in the agentic loop, not after the repo is on disk.
Sub-3s p95 · 12.7M US marks · 14 social platforms · free for five
What's available today
Why this, why now
Claude Code writes the scaffold. Etymolt clears the name.
The Claude Code agentic loop is end-to-end: a prompt, a plan, file writes, a commit, sometimes a deploy. The missing primitive in that loop is a verification step on the project name — the one piece of state that touches USPTO, RDAP, GitHub, npm, and your future launch announcement all at once. Etymolt sits in the loop before the commit lands.
Three install paths
JSON config, CLI command, or remote URL.
I · Canonical
.claude/settings.json
The documented MCP path. Drop the block in .claude/settings.json (project) or ~/.claude.json (user-wide). Reload Claude Code. Etymolt appears in the tool list.
// .claude/settings.json (project-scoped)
// or
// ~/.claude.json (user-scoped)
{
"mcpServers": {
"etymolt": {
"command": "npx",
"args": ["-y", "@etymolt/mcp-server"],
"env": {
"ETYMOLT_API_BASE": "https://api.etymolt.com"
}
}
}
}The block accepts the same shape in both files. Project-scoped travels with the repo.
II · Quick add
claude mcp add etymolt
The published Claude Code CLI command does the same registration as the JSON block. Idempotent — re-running does not duplicate.
# 1. Install Claude Code if you haven't yet
npm install -g @anthropic-ai/claude-code
# 2. Add Etymolt with one command — the published claude mcp add syntax
claude mcp add etymolt npx -y @etymolt/mcp-server
# 3. (Optional) Scope user-wide instead of per-project
claude mcp add --scope user etymolt npx -y @etymolt/mcp-serverNo API key required for the free tier. Five anonymous verdicts per IP are free.
III · Direct URL · remote MCP variant
# Claude Code also accepts remote MCP endpoints
# (HTTP transport). Functionally interchangeable with the stdio block.
claude mcp add --transport sse etymolt https://api.etymolt.com/mcpFor users who want HTTP transport instead of stdio. Functionally interchangeable; same verdict shape.
How it fires
Claude Code reaches for Etymolt on every naming intent.
I
You prompt Claude Code.
"npm init for a new SaaS", "scaffold a Rust crate for X", "rename this package" — any prompt that signals naming intent. Claude Code proposes the candidate inside the agentic loop.
II
Claude Code calls verify_brand_name via MCP.
Etymolt fans out across USPTO + RDAP + 14 social probes in parallel. Sub-3s p95. The agentic loop sees the verdict in the tool-result and routes accordingly.
III
Commit step uses the verdict.
PROCEED candidates land on disk. ABANDON candidates never get committed. ITERATE and DUE_DILIGENCE candidates surface with the blocker called out — the loop tries another name.
You Claude Code Etymolt (MCP) │ │ │ │ "npm init X" │ │ │ ───────────────────▶ │ │ │ │ naming intent fires │ │ │ ─────────────────────▶│ │ │ call: verify_brand_name │ │ │ │ │ ◀─── verdict + score ─┤ │ ◀── verdict + ── │ │ │ disclaimer │ │ │ (verbatim) │ │
The validation stack
Five axes. One signed verdict.
Trademark
USPTO · TTAB · UKIPO · Madrid
Domain & handles
RDAP · WHOIS · 14 platforms
Cultural
7 languages · 20 markets
Linguistic
1.4M-name distinctiveness corpus
Sound symbolism
12 perceptual axes · TTS round-trip
What we are NOT
We don't generate names. We validate them.
Claude Code's model proposes the candidate. Etymolt verifies it across five axes against live ground truth. The two products sit cleanly on different sides of the workflow.
FAQ
Common questions.
How do I install Etymolt in Claude Code?
Two paths. (1) Add the mcpServers block to .claude/settings.json (project-scoped) or ~/.claude.json (user-scoped) — both files accept the same shape. (2) Run claude mcp add etymolt npx -y @etymolt/mcp-server once. Either path registers Etymolt for every Claude Code prompt the agentic loop produces.
Where does Claude Code keep its config?
Claude Code reads two files. .claude/settings.json is project-scoped and travels with the repo — useful when a whole team wants the same MCP servers. ~/.claude.json is user-scoped and applies to every project on your machine. The mcpServers block has identical shape in both.
When does Claude Code actually call the tool?
Whenever the conversation matches naming intent — npm init, cargo new, package.json name fields, repo rename intents, project scaffolding inside the agentic loop. The MCP tool description fires on context, not friction-phrases, so Claude Code reaches for verify_brand_name before committing the project name.
Why register Etymolt for a Claude Code workspace?
Claude Code generates project names, package identifiers, deploy targets, and repo names — names the model has not actually verified against live USPTO, RDAP, or social-handle ground truth. Etymolt is the verification primitive that runs before the scaffold lands on disk. Five axes, sub-three-second p95.
Does this generate names? Does it compete with Claude Code?
No. Etymolt does not generate names. We validate the names Claude Code's model already proposes. Claude Code makes the candidates; we return the verdict. The two products sit cleanly on different sides of the workflow.
Does Etymolt see my prompts or generated code?
No. When Claude Code invokes verify_brand_name, only the name string and an optional category hint travel to api.etymolt.com. No prompt content, no generated code, no repo identifiers, no Anthropic account state. We log: name, timestamp, hashed source IP. Retained 90 days.
What does this cost?
Five verdicts per IP are free, no API key required. Signed-up users get ten more on the house. After that, ten dollars buys ten more. No subscription. The Claude Code seat itself is billed by Anthropic — separate line item.
Is there an Anthropic example skill or marketplace listing?
Yes — see sdks/etymolt-mcp/skills/naming-guard/SKILL.md in our public repo for the canonical naming-guard skill. Submission to the Anthropic Connectors Directory and the example-skills index is in flight per the 2026-05-18 board memo.
Bureau Model · clearance signal · not legal advice
Bring verdicts into Claude Code.
We don't generate names · we validate them · see all surfaces →
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.
NPM · @etymolt/mcp-server