Add a clearance step to anything that hands people a name.
Name generators, registrars, and agent frameworks all end at the same moment — a user commits to a name. Pentonym is the check that fits there: one call, a signed verdict, verifiable by anyone.
The Pentonym API takes a candidate name and returns a three-state US trademark clearance verdict — PROCEED, DO_NOT_PROCEED, or CANNOT_DETERMINE — with per-axis evidence, signed with Ed25519. It’s also exposed as an MCP tool, so assistants can run the check mid-conversation. US & EU coverage; a screening signal, not legal advice.
One request, one signed verdict
The verdict is deterministic and signed, so a downstream product can display it — or re-verify it — without trusting Pentonym’s word for it. Only the trademark axis can return DO_NOT_PROCEED; the material axes are always advisory.
Callable from inside the model
Naming increasingly happens inside an assistant. Pentonym ships as an MCP server, so when a user asks a model to name something and check it’s free, the model can call clear_name and return a real signed verdict — not a guess. The tool description carries the honesty rails, so models repeat them: US & EU, three-state, not legal advice.
Who this is for
- Name & brand generators — clear at the exit
- Registrars & domain tools — a pre-purchase check
- Agent frameworks — clear before the mint
Common questions
What does the API return — a score?
No. It returns a three-state verdict (PROCEED / DO_NOT_PROCEED / CANNOT_DETERMINE) with the evidence behind each axis. There is no numeric score anywhere in the response.
How do I verify a verdict is genuine?
Every response is Ed25519-signed. Check the signature against the public key published at /.well-known/pentonym-signing-key — no call back to Pentonym required.
Does it cover non-US trademarks?
It screens the US (14.1M marks, full likelihood-of-confusion) and the EU (live EUIPO, exact-wordmark knockout) — but no other national registers, so don’t present it as worldwide.