ClaimHit / Integrations
Drive it from your own tools. Or just ask Claude.
ClaimHit speaks the Model Context Protocol, so you can run a patent search from inside Claude and keep asking questions of the result without leaving the conversation. The same engine answers over REST, with a published OpenAPI spec and a sandbox key that costs nothing to integrate against.
ClaimHit, inside Claude
Add one URL to Claude and sign in with your ClaimHit account. Seven tools arrive with it, and from then on the patent work happens in the conversation: ask it to search a patent, then narrow, sort and interrogate what comes back, then chart the one that matters.
What makes it worth having
A search returns a list, and the useful part is the half-hour after it. The connector is built for that half-hour: reading, narrowing, filtering, sorting, paging and looking again at a vague result spend no credit and are not capped, over a result you have already paid for. You are not metered for thinking.
The two tools that spend ask first. Each quotes what it will cost and what it matched before anything runs, and nothing happens until you confirm — because a model that decides on its own initiative to try a variation should not be able to spend your credits to do it.
Getting connected
It is an OAuth 2.0 server, so there is no key to paste or store: you authorise it the way you would any other app, and ClaimHit sees your account rather than a shared secret. The endpoint, the full tool list and the worked prompts are on the connector page in the app.
Spends no credit, and is not capped
- Read a result. Narrow by company or technology, filter by confidence band, re-sort, page through the whole list, or ask for the evidence and per-element scores behind one Hit Score.
- Look again at one target. Hand it a vague result and say what you want instead — a specific model, the latest one, a capability, a market — and it returns real, current products with a reason for each.
- List your scans. Recent patents, dates and top results.
- Check credits. What is left on the account.
Spends a credit, and says so first
- Search a patent. The multi-model ensemble, returning ranked targets with evidence. 1 search credit.
- Generate a Hit Chart. Element by element against a chosen target. 1 chart credit.
- Re-run a search. Fresh results, repeating the inputs that steered the original. No credit within 6 months.
The same engine, from your own system
A REST surface at /api/v1, documented as OpenAPI 3.1 and served as a live reference you can read and try. Bearer authentication, typed enums, paginated envelopes — written to read the way your engineers expect an API to read.
How an assessment runs
Asynchronously, because a real one takes 60 to 180 seconds and returning it inline would make the first integration anybody writes a flaky one. You POST a patent, get a 202 and an id, then either poll it or give us a callback_url and we post the finished assessment to you.
A patent does not have to be published. Send a publication number, or send the claim text, or both — a field you supply wins and a field you leave out comes from the grant. That exists for the claim set that is being drafted or amended, which no number can reach.
An assessment finishes in phases, and the contract says which: the first pass lands in a couple of minutes and background enrichment keeps finding implementers for several more, so a payload names what is still outstanding rather than looking finished early. Act on it when it says COMPLETE.
Integrating against it
Sandbox keys begin ch_test_ and serve pre-seeded fixtures through the same code path as a live key — no engine call, nothing charged, and identical response shapes, because a sandbox whose shapes differ from production certifies an integration that breaks on the first real call. Live keys begin ch_live_. The rate limit is 20 requests a minute, published so a client that throttles correctly does not get 429s anyway.
What it covers
- Assessments. Start one for a patent, list them, read one with its targets and claim elements.
- Charts. Element-by-element against a chosen target from an assessment, up to three claims in one job.
- Portfolio. Push the patents an owner holds and the date that should trigger an assessment on each.
- Campaigns. Group targets from a portfolio scan into a licensing programme, move each one’s stage, close it out. State only — nothing here spends a credit.
- Account. What is left on the key.
Read the spec
The full reference is at api-docs — every path, every schema, every error code, and the shapes still open to review until somebody is in production behind them. Tell us what you need and it gets added: campaigns went in that way, as 1.1.0, without changing anything a client was already calling.
Two things that travel with the data
What comes back is research, not a determination. Every assessment and every chart carries a disclaimer saying so, and the contract requires it to be displayed wherever the data is shown — including inside your product. Results are candidates worth investigating, cross-checked against public evidence you can open; they are not a finding of infringement, essentiality or validity, and a qualified practitioner has to review them before anyone relies on them.
An API integration is a conversation, not a signup. Keys are issued per partner, scoped to an organisation and metered separately, so we talk first about volume, what you are building and which paths you actually need. The connector, by contrast, you can simply switch on with the account you already have.
Ask about API access
Tell us what you are integrating and roughly what volume you expect, and a person replies within one business day. The sandbox is available while you evaluate.
See who is using your patent.
The pilot is one patent: the full product search and one Hit Chart, with a 30-day money-back guarantee. No subscription, no card on file after that.
One patent, the full search and one Hit Chart. 30-day money-back guarantee.