Guide / Company data
How to find a company domain from its name in Wikidata and RDAP
Search Wikidata by legal name, take the official website only when one company matches, confirm the domain in RDAP, and flag the rest rather than guess.
Search Wikidata for "Stripe" and the first five results include a payments company, a band of colour, a Gremlins character and a family of beetles. That is the problem with turning a company name into a domain. Finding a domain is easy; knowing it belongs to the company you meant is the job.
Two free services do most of the work. Wikidata is a free, structured knowledge base from the Wikimedia Foundation, the nonprofit behind Wikipedia, and many company entries list an official website. RDAP, the Registration Data Access Protocol, is the modern replacement for WHOIS: ask it about a domain and the registry says whether it is registered, when it expires and which nameservers it uses.
By the end, your agent will turn a list of company names into domains it can stand behind, each tied to one company record and checked with the domain's registry, with every doubtful row flagged for a person. You need an HTTP client and, for each company, its legal name or one more fact about it, such as a country, an industry or an identifier.
How to find a company domain from its name
- Search Wikidata with
wbsearchentities, which searches labels and aliases, using the company's legal name (Stripe, Inc.) rather than its short name (Stripe). - Accept a match only when exactly one result is described as a company. With several candidates, search again with the second fact, or send the row to a person with the candidates attached; never pick by rank.
- Read the item's official website with
wbgetclaimsand P856, Wikidata's property for it, skipping any website with an end date. - Cut the URL down to its registrable domain:
https://stripe.com/becomesstripe.com, and Daikin'shttp://www.daikin.co.jpbecomesdaikin.co.jp, notco.jp. - Look the domain up at
https://rdap.org/domain/<domain>and confirm the record shows an expiration date in the future and at least one nameserver. - Write back the company, its Wikidata id, the domain, the date checked and one status: confirmed, ambiguous, not found or domain unverified.
Details that trip agents up
- No good match is an answer. If no result fits, mark the row not found and stop. A guessed domain does more damage than a blank one.
- Companies change websites. P856 keeps former websites beside the current one, so when it holds more than one URL, take the preferred-rank statement and skip any with an end time.
- Be a polite client. Send a descriptive User-Agent on every request, and when Wikidata answers 429, wait as long as its Retry-After header says. Agents with an AI client can add Wikidata's read-only MCP (Model Context Protocol) connector at
https://wd-mcp.wmcloud.org/mcp/instead. - rdap.org forwards, it doesn't answer. RDAP.org passes each query to the registry that is authoritative for the domain, so follow its redirect.
- Keep what you need, drop the rest. An RDAP record can carry contact entries. Confirming a domain needs none of them, so discard them.
- Registered isn't the same as working. Expiry and nameservers say the domain is registered and delegated; whether a page loads is a separate check.
Example calls
We sent the bare name to Wikidata's search with a limit of five and got five items back, only one of them a company. The response below is trimmed to the useful fields, and one entry about an individual is left out.
GET https://www.wikidata.org/w/api.php?action=wbsearchentities&language=en&type=item&format=json&limit=5&search=Stripe
{
"searchinfo": {"search": "Stripe"},
"search": [
{"id": "Q7624104", "label": "Stripe",
"description": "Irish-American payment technology company", ...},
{"id": "Q3421342", "label": "stripe",
"description": "long, narrow band of color, often in alternating sets", ...},
{"id": "Q127900502", "label": "Stripe",
"description": "fictional character in the Gremlins franchise", ...},
...
{"id": "Q297115", "label": "Meloidae", "description": "family of beetles",
"match": {"type": "alias", "language": "en", "text": "Striped Blister Beetle"}, ...}
],
"search-continue": 5,
"success": 1
}
The company ranked first, so taking the top hit would have worked this time. It is still the wrong rule: a smaller company named after an everyday word can't count on ranking first, and an agent can explain a match in a way it can never explain a rank.
The legal name returned a single item, the same company, and the response says why: it matched on the alias Stripe, Inc..
GET https://www.wikidata.org/w/api.php?action=wbsearchentities&language=en&type=item&format=json&limit=5&search=Stripe%2C%20Inc.
{
"searchinfo": {"search": "Stripe, Inc."},
"search": [
{"id": "Q7624104", "label": "Stripe",
"description": "Irish-American payment technology company",
"match": {"type": "alias", "language": "mul", "text": "Stripe, Inc."},
"aliases": ["Stripe, Inc."], ...}
],
"success": 1
}
Reading P856 on that item returned one statement, the URL https://stripe.com/, at normal rank and with no end time.
GET https://www.wikidata.org/w/api.php?action=wbgetclaims&property=P856&format=json&entity=Q7624104
{
"claims": {
"P856": [
{"mainsnak": {"snaktype": "value", "property": "P856",
"datavalue": {"value": "https://stripe.com/", "type": "string"},
"datatype": "url"},
"type": "statement", "rank": "normal", ...}
]
}
}
The RDAP lookup for stripe.com came back from Verisign's .com server as a domain object with four nameservers, an expiration date in September 2027 and six status values that prohibit deletion, transfer and updates. The response below omits the registrar entity and its contacts.
GET https://rdap.org/domain/stripe.com
{
"objectClassName": "domain",
"ldhName": "STRIPE.COM",
"links": [{"rel": "self",
"href": "https://rdap.verisign.com/com/v1/domain/stripe.com", ...}, ...],
"status": ["client delete prohibited", "client transfer prohibited",
"client update prohibited", "server delete prohibited",
"server transfer prohibited", "server update prohibited"],
"events": [
{"eventAction": "registration", "eventDate": "1995-09-12T04:00:00Z"},
{"eventAction": "expiration", "eventDate": "2027-09-11T04:00:00Z"},
...
],
"nameservers": [
{"objectClassName": "nameserver", "ldhName": "NS-1087.AWSDNS-07.ORG"},
...
],
...
}
The expiry check works beyond .com. ICANN, the body that coordinates the domain name system, requires every generic top-level domain's RDAP answer to carry registration and expiration dates under its RDAP response profile.
The record leaves out the owner: its only entity is the registrar. Wikidata ties the domain to the company, and RDAP shows the domain is registered and delegated. A row is confirmed only when both answer.
What it costs
Each of the four calls above cost $0. A clear name takes three calls: a search, a claims read and an RDAP lookup. An ambiguous one adds a second search.
| Call | Price per call | Limit to plan around |
|---|---|---|
Wikidata wbsearchentities |
$0 | Wikimedia's global rate limits; back off on 429 |
Wikidata wbgetclaims |
$0 | Same |
| RDAP via rdap.org | $0 | 10 requests in 10 seconds, then 429 |
Wikidata's data is CC0, a public-domain dedication, so no attribution is required, though Wikidata asks to be credited. The real limit is rdap.org's: only the RDAP lookup counts against it, one per name, which caps a run at about one name a second. RDAP.org says a client making that many queries should read the registry lists itself, and asks heavy users to support the service, which its maintainer runs at their own cost.
The cost that decides the method is the review queue. Every row marked ambiguous waits for a person, so count that share on your own list before you compare this path with a paid one.
When this is the wrong tool
Very large lists. Wikidata's guidance says the Action API is a poor fit for large result sets and points to its database dumps, which are not current because they take a long time to export.
Companies Wikidata does not hold. Wikidata holds over 120 million items, but small private firms are the likeliest to be missing, and each one becomes a not-found row.
Country-code domains. Only some country-code registries, the ones behind two-letter endings for countries, follow the ICANN profile. A 404 from rdap.org means it knows no authoritative RDAP server for that domain, not that the domain is unregistered, so mark those rows domain unverified.
Proof of ownership. RDAP may withhold who registered a domain, and the stripe.com record named only the registrar. If you need to prove the company holds the domain, this path does not provide that.
Domains, not names. If the agent holds domains or work-email addresses and needs the company behind them, our domain-to-company guide runs the reverse.
When the agent needs firmographics anyway. A paid enrichment API may resolve the domain as part of the company record. Judge it on match rate, freshness and cost per enriched record, and ask what its match rate counts as the denominator.
Frequently asked questions
Is there a free API to get a company's website from its name?
Wikidata's search API finds the company's item, and its official website property (P856) holds the URL. Both are free to call, and the data is CC0, a public-domain dedication.
Why does searching a company name return the wrong entity?
Wikidata's search matches labels and aliases, so a short name like Stripe also matches a band of colour, a fictional character and a beetle family. Searching the legal name, or adding a second fact, narrows it to one item.
Does an RDAP lookup prove a company owns a domain?
No. RDAP confirms a domain is registered, shows when it expires and lists its nameservers, but registrant details may be withheld, so the link to the company has to come from another source such as Wikidata.
What does a 404 from rdap.org mean?
It means rdap.org knows no authoritative RDAP server for that domain, which happens with some country-code domains. Treat the domain as unverified, not as unregistered.
Can an AI agent query Wikidata directly?
Yes. Wikidata runs a read-only MCP (Model Context Protocol) server at https://wd-mcp.wmcloud.org/mcp/ that can be added as a connector in an AI client, and the plain Action API works from any HTTP client.
Read this article as Markdown Open in ChatGPT Open in Claude


