Inscriptions inscribe sats with arbitrary content, creating bitcoin-native digital artifacts, more commonly known as NFTs. Inscriptions do not require a sidechain or separate token.

These inscribed sats can then be transferred using bitcoin transactions, sent to bitcoin addresses, and held in bitcoin UTXOs. These transactions, addresses, and UTXOs are normal bitcoin transactions, addresses, and UTXOS in all respects, with the exception that in order to send individual sats, transactions must control the order and value of inputs and outputs according to ordinal theory.

The inscription content model is that of the web. An inscription consists of a content type, also known as a MIME type, and the content itself, which is a byte string. This allows inscription content to be returned from a web server, and for creating HTML inscriptions that use and remix the content of other inscriptions.

Inscription content is entirely on-chain, stored in taproot script-path spend scripts. Taproot scripts have very few restrictions on their content, and additionally receive the witness discount, making inscription content storage relatively economical.

Since taproot script spends can only be made from existing taproot outputs, inscriptions are made using a two-phase commit/reveal procedure. First, in the commit transaction, a taproot output committing to a script containing the inscription content is created. Second, in the reveal transaction, the output created by the commit transaction is spent, revealing the inscription content on-chain.

Inscription content is serialized using data pushes within unexecuted conditionals, called "envelopes". Envelopes consist of an OP_FALSE OP_IF … OP_ENDIF wrapping any number of data pushes. Because envelopes are effectively no-ops, they do not change the semantics of the script in which they are included, and can be combined with any other locking script.

A text inscription containing the string "Hello, world!" is serialized as follows:

  OP_PUSH "ord"
  OP_PUSH "text/plain;charset=utf-8"
  OP_PUSH "Hello, world!"

First the string ord is pushed, to disambiguate inscriptions from other uses of envelopes.

OP_PUSH 1 indicates that the next push contains the content type, and OP_PUSH 0indicates that subsequent data pushes contain the content itself. Multiple data pushes must be used for large inscriptions, as one of taproot's few restrictions is that individual data pushes may not be larger than 520 bytes.

The inscription content is contained within the input of a reveal transaction, and the inscription is made on the first sat of its input. This sat can then be tracked using the familiar rules of ordinal theory, allowing it to be transferred, bought, sold, lost to fees, and recovered.


The data model of inscriptions is that of a HTTP response, allowing inscription content to be served by a web server and viewed in a web browser.


Inscriptions may include fields before an optional body. Each field consists of two data pushes, a tag and a value.

Currently, there are six defined fields:

  • content_type, with a tag of 1, whose value is the MIME type of the body.
  • pointer, with a tag of 2, see pointer docs.
  • parent, with a tag of 3, see provenance.
  • metadata, with a tag of 5, see metadata.
  • metaprotocol, with a tag of 7, whose value is the metaprotocol identifier.
  • content_encoding, with a tag of 9, whose value is the encoding of the body.

The beginning of the body and end of fields is indicated with an empty data push.

Unrecognized tags are interpreted differently depending on whether they are even or odd, following the "it's okay to be odd" rule used by the Lightning Network.

Even tags are used for fields which may affect creation, initial assignment, or transfer of an inscription. Thus, inscriptions with unrecognized even fields must be displayed as "unbound", that is, without a location.

Odd tags are used for fields which do not affect creation, initial assignment, or transfer, such as additional metadata, and thus are safe to ignore.

Inscription IDs

The inscriptions are contained within the inputs of a reveal transaction. In order to uniquely identify them they are assigned an ID of the form:


The part in front of the i is the transaction ID (txid) of the reveal transaction. The number after the i defines the index (starting at 0) of new inscriptions being inscribed in the reveal transaction.

Inscriptions can either be located in different inputs, within the same input or a combination of both. In any case the ordering is clear, since a parser would go through the inputs consecutively and look for all inscription envelopes.

InputInscription CountIndices
02i0, i1
23i3, i4, i5


HTML and SVG inscriptions are sandboxed in order to prevent references to off-chain content, thus keeping inscriptions immutable and self-contained.

This is accomplished by loading HTML and SVG inscriptions inside iframes with the sandbox attribute, as well as serving inscription content with Content-Security-Policy headers.