APIShed MCP

Computes the WCAG 2.1 contrast ratio between two hex colors. Alpha, if present, is composited over white first.

Converts a color between hex, rgb, hsl, and hsv. value's shape depends on from/to: a string for hex, an object ({r,g,b,a?} etc.) otherwise.

Converts JSON records to CSV text. columns is the explicit, ordered header — JSON object key order can't be relied on to infer it.

Parses CSV text into rows of string fields (no type guessing). Ragged rows are tolerated — use csv_validate to detect them.

Parses CSV text (header row required) and reports per-column non-empty/empty value counts across the data rows. No numeric aggregation.

Converts CSV text to JSON: header=true (default) returns one object per data row keyed by the header row; header=false returns raw rows. Every value is a string.

Reports whether CSV text is structurally well-formed: every row has the same field count as the first.

Adds years/months/days to date. years/months are applied together with end-of-month clamping — e.g. 2026-01-31 + 1 month = 2026-02-28, not an overflowed 2026-03-03. days is then applied as plain calendar-day arithmetic.

Age in whole years as of asOf (optional, defaults to the current UTC date). A Feb-29 birthDate is treated as falling on March 1 in a non-leap asOf year.

Counts days in [from, to] (inclusive on both ends) whose weekday is in weekdays (default mon-fri) and which aren't in holidays.

Day of the week for date: both the weekday name and its ISO 8601 weekday number (1=Monday..7=Sunday).

Day difference between two YYYY-MM-DD dates: to minus from, in whole days. Negative if to is before from.

Checks whether year is a leap year in the Gregorian calendar.

Returns the actual current date/time — an LLM has a training cutoff and no live clock, so it can't know what time it actually is right now.

Validates an 8-digit (EAN-8), 12-digit (UPC-A), 13-digit (EAN-13), or 14-digit (GTIN-14) barcode checksum: alternating weights 1 3 applied right-to-left, check digit is (10 - sum mod 10) mod 10.

Checks value is a single, bare RFC 5322 email address — syntax only, not a deliverability check. user@localhost is valid (no TLD required). "Name <email>" display syntax is rejected.

Decodes rot13, base64, base64url, base32, hex, uuencode, urlcomponent, urlform, or html text back to the original. encoding in the result is "text" if the decoded bytes are valid UTF-8, or "base64" if they're binary.

Encodes plain UTF-8 text into rot13, base64, base64url, base32, hex, uuencode, urlcomponent (RFC 3986 percent-encoding, space -> %20), urlform (application/x-www-form-urlencoded, space -> +), or html (entity-escapes < > & ' "). Returns plain ASCII.

Computes a hash digest of input (decoded per encoding: "text" default, "base64", or "hex") using algo: md5, sha1, sha256, sha512, or crc32. Returns the digest as lowercase hex.

Computes an HMAC of input (decoded per encoding) using algo: md5, sha1, sha256, or sha512 (crc32 isn't offered — it's a checksum, not a cryptographic primitive HMAC composes with). key is used as its raw UTF-8 bytes, not decoded per encoding. Returns the MAC as lowercase hex.

Validates an IBAN's length for its issuing country (per the SWIFT IBAN registry) and its mod-97 checksum. Does not decompose the BBAN into a bank code/account number or resolve a BIC.

Validates a 10- or 13-character ISBN. ISBN-10 uses weights 10..1 over 10 characters (last may be X, worth 10), sum mod 11 must be 0. ISBN-13 uses the same GTIN checksum as barcodes.

Validates document against schema (JSON Schema draft-07 or 2020-12). Reports one specific violation per call, not an aggregated list.

Validates a numeric string against the Luhn (mod-10) checksum — payment card numbers, IMEI numbers, and similar identifiers.

Reports whether a CIDR block contains an address. Different address families (IPv4 vs IPv6) never match.

Parses a CIDR block and reports its network address, address range, and address count. An address with host bits set (e.g. 10.14.32.17/21) is normalized to its network address in the result.

Reports whether two CIDR blocks share any address.

Validates the checksum of a 10-digit Polish tax identification number (NIP). Checksum uses weights 6 5 7 2 3 4 5 6 7 over the first 9 digits; sum mod 11 must equal the 10th digit. A sum mod 11 of 10 is itself invalid.

Validates the checksum of an 11-digit Polish national identification number (PESEL). Checksum uses weights 1 3 7 9 1 3 7 9 1 3 over the first 10 digits; control digit is (10 - sum mod 10) mod 10. On success also decodes the date of birth and sex embedded in the number; a checksum-valid number with an impossible calendar date (e.g. Feb 30) is still reported invalid.

Runs a short ordered chain of existing apished operations in one call (same as POST /v1/pipe/run), passing one named field of each step's result into the next step's args. Stops at the first failing step.

Generates cryptographically random integers in [min, max] (inclusive), using crypto/rand — an actual random value, not a guess. count defaults to 1, max 1000.

Extracts up to maxMatches non-overlapping regex matches from input, using Go's RE2-based regexp engine (linear-time, no backreferences/lookaround). Named groups (?P<name>...) are reported in namedGroups.

Tests input against pattern (Go's regexp package: RE2 syntax, not PCRE — no backreferences or lookaround; use inline flags like (?i) for case-insensitive matching). Returns the full match and captured groups.

Replaces every regex match of pattern in input with replacement, which may reference capture groups as $1, $name, etc. (Go's Regexp.Expand syntax).

Splits input around regex matches of pattern, returning at most limit substrings (the last holds the unsplit remainder). limit<=0 means no limit.

Validates the checksum of a 9- or 14-digit Polish business registry number (REGON). 9-digit: weights 8 9 2 3 4 5 6 7 over the first 8 digits, sum mod 11 (10 mapped to 0) must equal the 9th digit. 14-digit: the first 9 digits must themselves be a valid 9-digit REGON, and weights 2 4 8 5 0 9 7 3 6 1 2 4 8 over the first 13 digits (sum mod 11, 10 mapped to 0) must equal the 14th digit.

Evaluates a Reverse Polish Notation (postfix) expression: whitespace-separated tokens, each a number or an operator/function that pops its operands off a stack and pushes the result. Supports arithmetic, exponentiation/modulo, roots, trigonometry, logarithms, rounding, bitwise ops, and constants (pi, e, phi). Always returns the decimal result — for hex/bin/oct rendering, convert the returned numbe

Structural format validation only for a SWIFT/BIC code: 8 or 11 characters (4 letters bank code + 2 letters country code + 2 alphanumeric location code + optional 3 alphanumeric branch code). The country code is not cross-checked against ISO 3166-1.

Returns the 0-indexed character (rune) at index in text.

Counts non-overlapping occurrences of substring in text (rune-safe, not byte-based).

Checks whether text reads the same forwards and backwards. ignoreCase/ignoreNonAlnum default false (strict, literal comparison); set both true for phrase-style palindromes like "A man, a plan, a canal: Panama".

Reverses text by Unicode code point (rune), not grapheme cluster.

Sorts text's characters or words alphabetically. by: "chars" (default, sorts runes, no separator) or "words" (splits on whitespace, rejoins with a single space).

Descriptive statistics for text: rune/byte/word/line counts, a unique-rune count, a letters/digits/spaces/punctuation/other breakdown, an exact per-character frequency table, and word-length min/max/average plus the actual shortest/longest word (ties resolve to the first occurrence).

Counts whitespace-delimited words in text.

Counts tokens under an OpenAI-compatible (tiktoken) encoding. Provide exactly one of model (e.g. "gpt-4o") or encoding ("o200k_base" or "cl100k_base"). No Claude/Anthropic tokenizer.

Converts a data-size value (decimal string) between bit, Kbit, Mbit, Gbit, Tbit (SI bits), B, KB, MB, GB, TB, PB (SI bytes), and KiB, MiB, GiB, TiB, PiB (IEC bytes). Always exact.

Converts a temperature value (decimal string) between C, F, and K. Offset-aware, not a bare multiplier.

Checks value is a well-formed absolute URL (has both a scheme and a host) — syntax only, not a reachability check.

Generates count (default 1, max 1000) random version-4 UUIDs from crypto/rand-backed entropy.

Parses value as a UUID and reports its version/variant if well-formed.

Format validation only (no VIES lookup) for an EU VAT number: checks the 2-letter member-state prefix (EL for Greece, XI for Northern Ireland) against that country's digit/letter pattern. A format-valid result does not mean the number is actually registered.