One request, ranked conflicts
Screen “nikee” for apparel (international class 25):
curl "https://api.markbase.co/clearance?mark=nikee&class=25"The response ranks existing marks by similarity, each with a score and risk level:
{
"mark": "nikee",
"class": "025",
"highest_risk": "high",
"matches": [
{
"serial_number": "72414177",
"word_mark": "NIKE",
"owner_name": "Nike, Inc.",
"status_code": "800",
"international_codes": ["025"],
"similarity": 0.85,
"risk": "high"
}
]
}Use highest_risk to prioritize review. It reports spelling-based risk, not a legal availability decision or a phonetic probability.
Add whole-name phonetic evidence
Add phonetic=true to the single-name request, or set the JSON boolean phonetic to true in a batch. Node SDK 2026.11.2 and later accept phonetic: true in either method's options. Omit the option or use false to keep spelling-only behavior. Quota remains one monthly request per name.
Opt-in results add matching metadata and separate signals.exact, signals.spelling.score, and signals.phonetic evidence. Double Metaphone compares ordered ASCII alphabetic tokens across the whole name. Changed tokens need at least four letters; shorter tokens must match literally in the same position. Names with digits, non-ASCII characters, more than 12 tokens, or only short tokens return not_evaluated while retaining spelling results.
Exact matches rank first, followed by spelling scores of at least 0.85, phonetic matches, and weaker spelling matches. similarity and risk retain their existing spelling meaning. highest_risk considers retrieved, verified candidates before the result limit, so it may describe a candidate outside the returned matches.
Show matching.truncated when retrieval reaches its candidate cap or hydration budget. A value of false does not mean the search was exhaustive. An unavailable phonetic request returns HTTP 503 without switching modes. Admission failures do not use monthly quota; an accepted batch still counts if a search fails. Avoid automatic batch retries.
Why class scoping matters
Trademark rights are tied to goods and services. DELTA coexists as an airline, a faucet maker, and dental insurance because they occupy different classes. Screening without a class produces noise; passing class=25 (or whichever of the 45 international classes fits your use) keeps results to marks that could actually block you. The /classes endpoint lists them all.
Where it fits
Name generators — screen every AI-suggested name before showing it, so users never fall in love with a taken mark.
Domain registrars — flag likely trademark conflicts at search time, before the customer buys a domain they can't safely use.
Legal intake — pre-qualify clearance requests so attorneys start from a candidate list instead of a blank search box.
AI agents — a deterministic tool call that grounds “is this name available?” in the actual USPTO register instead of a model's guess.
What it is — and isn't
The endpoint is a fast first-pass filter over the federal register. It doesn't cover state trademarks, common-law rights, or the legal analysis of confusion — no API does. Treat a clean result as “worth pursuing,” not “cleared,” and put a real attorney at the end of the funnel for names you intend to use.
Screen a name right now
No key needed on the free tier — try your own brand name against the live register.
curl "https://api.markbase.co/search?q=your-brand"Frequently asked questions
What is a trademark clearance search?
Before adopting a brand name, you check whether confusingly similar marks already exist for related goods or services. Attorneys do a deep version of this; the clearance API automates the first pass so obviously conflicted names die early.
Is this legal advice?
No. The API is a screening signal that surfaces similar existing marks. Likelihood-of-confusion is a legal judgment that depends on factors an algorithm can't weigh — for a name you're serious about, have a trademark attorney run a full clearance.
How is similarity scored?
The default search ranks retrieved US federal marks by spelling similarity. Opt-in phonetic matching adds whole-name sound-alike evidence and changes result ordering, while similarity and risk retain their spelling-based meaning. Class filters narrow the search; they do not resolve whether goods or services are legally related.
Can I screen names in bulk?
Yes. Launch and Scale keys can submit up to 25 names with POST /clearance/batch. Each name counts as one monthly request. Open and Free users can try one name at a time with GET /clearance.