On events of kind `0` (`set_metadata`) one can specify the key `"nip05"` with an [internet identifier](https://datatracker.ietf.org/doc/html/rfc5322#section-3.4.1) (an email-like address) as the value. Although there is a link to a very liberal "internet identifier" specification above, NIP-05 assumes the `<local-part>` part will be restricted to the characters `a-z0-9-_.`, case insensitive.
Upon seeing that, the client splits the identifier into `<local-part>` and `<domain>` and use these values to make a GET request to `https://<domain>/.well-known/nostr.json?name=<local-part>`.
The result should be a JSON document object with a key `"names"` that should then be a mapping of names to hex formatted public keys. If the public key for the given `<name>` matches the `pubkey` from the `set_metadata` event, the client then concludes that the given pubkey can indeed be referenced by its identifier.
A client may implement support for finding users' public keys from _internet identifiers_, the flow is the same as above, but reversed: first the client fetches the _well-known_ URL and from there it gets the public key of the user, then it tries to fetch the kind `0` event for that user and check if it has a matching `"nip05"`.
### Clients must always follow public keys, not NIP-05 addresses
For example, if after finding that `bob@bob.com` has the public key `abc...def`, the user clicks a button to follow that profile, the client must keep a primary reference to `abc...def`, not `bob@bob.com`. If, for any reason, the address `https://bob.com/.well-known/nostr.json?name=bob` starts returning the public key `1d2...e3f` at any time in the future, the client must not replace `abc...def` in his list of followed profiles for the user (but it should stop displaying "bob@bob.com" for that user, as that will have become an invalid `"nip05"` property).
A client can also use this to allow users to search other profiles. If a client has a search box or something like that, a user may be able to type "bob@example.com" there and the client would recognize that and do the proper queries to obtain a pubkey and suggest that to the user.
Clients may treat the identifier `_@domain` as the "root" identifier, and choose to display it as just the `<domain>`. For example, if Bob owns `bob.com`, he may not want an identifier like `bob@bob.com` as that is redundant. Instead, Bob can use the identifier `_@bob.com` and expect Nostr clients to show and treat that as just `bob.com` for all purposes.
### Reasoning for the `/.well-known/nostr.json?name=<local-part>` format
By adding the `<local-part>` as a query string instead of as part of the path the protocol can support both dynamic servers that can generate JSON on-demand and static servers with a JSON file in it that may contain multiple names.
JavaScript Nostr apps may be restricted by browser [CORS][] policies that prevent them from accessing `/.well-known/nostr.json` on the user's domain. When CORS prevents JS from loading a resource, the JS program sees it as a network failure identical to the resource not existing, so it is not possible for a pure-JS app to tell the user for certain that the failure was caused by a CORS issue. JS Nostr apps that see network failures requesting `/.well-known/nostr.json` files may want to recommend to users that they check the CORS policy of their servers, e.g.:
Users should ensure that their `/.well-known/nostr.json` is served with the HTTP header `Access-Control-Allow-Origin: *` to ensure it can be validated by pure JS apps running in modern browsers.
