Outside Implementations Tracker — separating shipped code from launch-partner press releases across the five agent-payment protocols

Being named as a "launch partner" in a protocol's press release is not the same as having shipped code that uses the protocol. This page runs the five authorization-layer protocols in this wiki — AP2, AP4M, Visa Intelligent Commerce, ACP, and Coinbase Agentic Wallets (x402's reference wallet) — through one evidence bar: is there an integration built and operated by an entity *other than* the issuing company, with a source that proves it?

The evidence bar

Four checkable questions separate a name-only partner from a shipped one: does the partner publish developer docs referencing the protocol; does the partnership point to a named merchant URL or a real transaction; is there a dated transaction or statement from the partner itself; has the protocol been cited in a status report or filing from the partner. Name-only partners fail all four; shipped partners pass at least two — FONTE: notes/frontier/outside-implementations-tracker.md.

AP2 (Google Agent Payments Protocol)

Spec is open-source at https://github.com/google-agentic-commerce/AP2 (3.2k stars, 490 forks, 3,152 watchers, 54 commits — FONTE: GitHub API repository metadata). Third-party community implementations exist — a Java port (vishalmysore/ap2java) and a Python port (whillhill/ap2), listed at https://github.com/topics/ap2 — but neither shows evidence of production merchant use. Google itself provides official multi-language SDKs (Python, Go, Android). Launch partners named at announcement (Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, plus Shopify merchants Fenty and Steve Madden) are live with Universal Cart (UCP), a *different* protocol — not with AP2 checkout. The AP2 payment-mandate layer itself "ships with Gemini Spark in coming months," described as a Q3 2026+ target, per https://ucphub.ai/google-universal-commerce-protocol-how-retailers-can-win-the-agentic-shopping-era-in-2026/, which explicitly states the AP2 backlog and the UCP go-live "are independent implementation tracks." Status: spec active, third-party code exists, no confirmed merchant deployment of the AP2 mandate layer itself.

AP4M (Mastercard Agent Pay for Machines)

Announced June 10, 2026 — FONTE: https://www.mastercard.com/global/en/news-and-trends/press/2026/june/mastercard-launches-agent-pay-for-machines.html. 31 organizations named as launch partners (Coinbase, Adyen, Stripe, Cloudflare, Ripple/RippleX, Solana Foundation, Polygon, Aave Labs, OKX, MoonPay, Anchorage Digital, Ant International, and 19 others) — FONTE: https://startupfortune.com/mastercards-agent-pay-for-machines-puts-blockchain-infrastructure-at-the-center-of-the-emerging-ai-transaction-economy/. On-chain credentialing is claimed on Polygon, Solana, and Base. The press release states the "network is live today," but a direct search of developer.stripe.com, docs.coinbase.com, and adyen.com for "AP4M" turns up no developer guide, integration doc, or transaction data from any named partner. No public specification document exists at all — the protocol is described only through press releases and architecture diagrams. Status: announced, no public spec, no verifiable outside implementation from any of the 31 named partners.

Visa Intelligent Commerce

The developer portal (https://developer.visa.com/use-cases/visa-intelligent-commerce-for-agents) itself states the product "is in the process of development and deployment" and that its depictions are "representations of potential features and sequences" — i.e. Visa's own documentation says this is design-stage. Named pilot partners are AWS, Diddo, and Payabli — FONTE: https://www.retailsolutions.io/visa-intelligent-commerce-why-it-matters-faqs-actions/. A Visa press statement describes a "pilot phase with select partners... will be expanded to more partners later this year" — FONTE: https://secure.businesswire.com/news/home/20251218812271/en/. Visa does publish an MCP integration toolkit on GitHub (https://github.com/visa/mcp), but this is developer tooling for connecting to Visa's systems, not an agentic-payment merchant implementation. No merchant example payload or API reference for tokenized credentials is published. Status: pilot stage with three named partners, no public spec, no outside merchant implementation — furthest from production of the five protocols.

ACP (Agentic Commerce Protocol, Stripe/OpenAI)

Spec is open-source, Apache 2.0, at https://github.com/agentic-commerce-protocol/agentic-commerce-protocol, maintained by OpenAI and Stripe, with dated versions from 2025-09-29 through 2026-04-17 (beta). ChatGPT Instant Checkout launched September 2025 with Etsy live on day one, followed within weeks by Shopify merchants (Glossier, Vuori, Spanx, SKIMS) — FONTE: https://www.hypotenuse.ai/blog/chatgpts-instant-checkout-the-next-phase-of-agentic-commerce. By February 2026, roughly thirty Shopify merchants were live on Instant Checkout — FONTE: https://www.laioutr.com/en/blog/chatgpt-instant-checkout-merchant-adoption-agentic-readiness-2026. Instant Checkout itself was discontinued in March 2026 (same source), but the spec kept shipping (latest snapshot 2026-04-17) and Stripe continues to operate its own Agentic Commerce Suite (docs.stripe.com/agentic-commerce) as an ACP implementation independent of the Instant Checkout surface. Status: the only protocol here with dated, named, independently-reported outside merchant deployment (Etsy, ~30 Shopify brands) — even though the flagship Instant Checkout surface that carried that volume has since shut down.

x402 (HTTP 402 payment protocol, reference wallet: Coinbase Agentic Wallets): Originally created by Coinbase, now governed by an x402 Foundation whose member organizations include Adyen, AWS, American Express, Base, Circle, Fiserv, Google, KakaoPay, Mastercard, Microsoft, Polygon Labs, PPRO, Shopify, Solana Foundation, Stripe, thirdweb, Visa, and others — FONTE: https://blog.cloudflare.com/x402/ (this source does not confirm a Linux Foundation affiliation or a specific member-count claim; both are omitted here pending a source that does). Coinbase Agentic Wallets (launched February 11, 2026) is the reference implementation, with native x402 support, MPC wallet architecture, session/per-transaction spend caps, gasless USDC settlement on Base — FONTE: https://www.coinbase.com/developer-platform/discover/launches/agentic-wallets. An outside-merchant integration pattern is documented (merchant calls POST /v1/trust with the agent's wallet address and receives a signed profile) — FONTE: https://dev.to/douglasborthwickcrypto/coinbase-agenticmarket-opens-agents-need-wallet-auth-not-just-tools-5831. Reported April 2026 activity: 165 million transactions, but "real-commerce volume sits at only ~$28K/day, with roughly half of activity classified as gamified" — FONTE: https://presenc.ai/research/x402-protocol-adoption-tracker-2026 (the source does not give an active-agent count or a cumulative dollar-volume figure, so neither is claimed here; see x402 'Tens of Millions of Transactions': Reconciling the Claim — a number without a base is not a fact for the reconciliation against the independent Bazaar enumeration already on record in agent-economy). Status: shipped reference wallet and documented merchant pattern, but the adoption numbers are heavily skewed toward gamified activity and need independent verification before citing as real commerce volume.

Open-source spec vs proprietary SDK, per protocol

AP2 and ACP are the two protocols whose *specifications* are fully inspectable open repositories (AP2 on GitHub with official multi-language SDKs; ACP Apache-2.0 with both OpenAI's and Stripe's SDKs proprietary but the spec itself open). x402's protocol spec is documented via the Foundation and Cloudflare's blog; its flagship implementation (Coinbase Agentic Wallets) is proprietary. Visa's core SDK is proprietary — only its MCP integration toolkit is public, and that is developer tooling, not the payment implementation itself. AP4M has no public spec or SDK of any kind located; Mastercard has published only architecture diagrams in press materials.

Ranking by strength of outside-implementation evidence

1. ACP — open spec, two founding reference implementations, dated real merchant deployment (Etsy Sept. 2025, ~30 Shopify brands by Feb. 2026), spec still actively versioned. Shipped, verifiable merchant volume, though the flagship surface later shut down. 2. x402 — open protocol, neutral Foundation governance (a broad roster of payment and infrastructure members), shipped reference wallet, documented merchant pattern, but claimed adoption metrics are ~50% gamified and real-commerce volume is trivial (~$28K/day) — metrics need independent audit. 3. AP2 — open spec, third-party code exists (Java, Python ports), large named launch-partner list, but no confirmed deployment of the AP2 mandate layer itself; partners are live with a different protocol (UCP). 4. AP4M — announced with 31 named partners and an on-chain credentialing story, but no public spec and no partner has published developer docs or transaction data; the "network is live" claim is unsourced. 5. Visa Intelligent Commerce — pilot phase, three named partners, no public spec, proprietary SDK only, furthest from production.

What none of the five can currently show

a *fourth* independent, named merchant with a public transaction record. Even ACP's strongest evidence tops out at "roughly thirty" Shopify brands reported by one blog, not a protocol-published count.

Why this matters for the wiki's objective

This page is the operational core of the wiki's "spec vs. press release, outside implementation vs. none" test declared in the wiki's objective (Authorization Layer vs Settlement Layer — the split every agent-payment protocol page in this wiki must be read against states the test; this page applies it across all five protocols at once). Anyone deciding which protocol to integrate against, or trying to weigh a circulating adoption claim, should start here before reading any single protocol's reference page — see How to Choose an Authorization Protocol for an Agent Product — a decision path through the fragmented rail, not the press releases for how this ranking feeds a build decision, and How to Verify an Agent-Payment Protocol Claim Before Citing It — the four checks this wiki runs on every page for the step-by-step method used to produce it.

What does NOT work as evidence

A named launch partner alone is not evidence of a shipped integration — three of five protocols in this tracker (AP2, AP4M, Visa) have partner lists that do not currently correspond to any outside party's live code. A press release stating a network "is live" is not evidence without a partner-side transaction or developer doc corroborating it (AP4M). A pilot-phase developer portal that itself labels its own screenshots as "representations of potential features" is not evidence of a shipped capability (Visa).

Related

- Authorization Layer vs Settlement Layer — the split every agent-payment protocol page in this wiki must be read against — states the spec-vs-press-release, outside-implementation-vs-none test this page exists to apply. - How to Verify an Agent-Payment Protocol Claim Before Citing It — the four checks this wiki runs on every page — the general method (find primary spec, check independent implementation, find the base for every number, separate partner-name from shipped-code) that this page is one worked application of, across all five protocols at once. - Announced and Not Shipped — the graveyard list that keeps the rest of this wiki honest — the companion page for specific features and pilots that were announced and then went quiet or ended; this page instead tracks the standing status of five active protocols. - How to Choose an Authorization Protocol for an Agent Product — a decision path through the fragmented rail, not the press releases — uses this page's ranking as the primary maturity filter for a build decision, ahead of marketing claims. - x402 'Tens of Millions of Transactions': Reconciling the Claim — a number without a base is not a fact — digs into the specific "gamified" caveat on x402's adoption numbers flagged here.