Agents that enrich leads or merge data sets hit the same wall: the same company shows up as "IBM Corp" and "I.B.M. Corporation, Inc.", the same person as "Bob Smith" and "Smith, Robert", the same address as "123 Main St Apt 4" and "123 Main Street #4". Matching those by hand does not scale, and the existing matching APIs sit behind signups and per-lookup pricing.
Matchpoint exposes the matching math directly. Each endpoint takes one messy value and returns a stable match key: normalize it, drop the noise (legal suffixes, punctuation, nicknames, formatting), phonetic-encode what is left, and hand back a key that collides for values meaning the same thing. Bucket your records by key and the duplicates fall together.
The cluster ships with seven endpoints, all pure compute with no upstream service behind them:
match-key-company— company and organization namesmatch-key-name— person names, with nickname and ordering handlingmatch-key-address— USPS-style address standardizationaddress-parse— split a freeform US address into number, street, unit, city, state, and zipmatch-key-email— Gmail dot and plus-tag collapse, domain canonicalizationmatch-key-phone— digits to canonical E.164match-score— compare two values of a given type and get a 0 to 100 similarity score plus whether they share a key
The pattern most agents will use: call a match-key-* endpoint to bucket
records, then call match-score to rank the candidate pairs inside a bucket by
confidence. Keys are deterministic, so the same input always returns the same
key — safe to cache and safe to compare across runs.
Matchpoint is the first cluster built straight from on-chain demand signal: our top x402 payers were already paying other providers for exactly this work, so we built the primitives they were reaching for.