Mastercard Start Path Names 22 Startups for Its First Agentic Commerce Cohort: Reading the Layers of Agent Checkout, Catalogs and Identity
Mastercard's Start Path has confirmed 22 companies for its new Agentic Commerce & Services track. We sort the cohort into checkout, catalog, identity and fraud layers, relate it to Agent Pay and UCP, compare it with Visa and Stripe, and explain which layer merchants should build first.
Key Takeaways
- Mastercard's startup engagement program Start Path confirmed in September 2026 the first 22-company cohort of its Agentic Commerce & Services track, a track it created in January 2026
- The 22 companies fall into layers (selling to agents and checkout, catalog optimization, agent identity, fraud and audit, enterprise deployment) that fill the merchant-side connections Agent Pay and UCP do not cover
- For merchants, the substance is that the layer that makes catalogs readable by agents and the layer that accepts agent-driven orders are now available as products, and the catalog comes first
A track created in January reached its first cohort in eight months
Mastercard's Start Path has selected 22 companies for its first Agentic Commerce & Services cohort, spanning payments and AI infrastructure.
thepaypers.comOn September 2, 2026, The Paypers reported that Start Path, Mastercard's startup engagement program, had admitted 22 companies as the inaugural cohort of its Agentic Commerce & Services track. Mastercard's official Start Path news page carries the same announcement under the heading "Start Path welcomes 22 new companies into its inaugural Agentic Commerce & Services program", with a one-line description of each company.
The track itself is new. On January 20, 2026, the week after NRF 2026, Mastercard published "The new rules of the road for agentic commerce", announcing at once that it had joined Google's Universal Commerce Protocol (UCP), that it was working with Microsoft to bring Agent Pay to Copilot Checkout, and that it was expanding Start Path into agentic commerce. The first cohort was confirmed a little under eight months later.
Start Path has taken in more than 500 companies from over 60 countries since it started in 2014. Agentic Commerce & Services is its seventh track, following acceptance, blockchain and digital assets, emerging fintech, open finance, small business, and security solutions. According to PYMNTS' January coverage, Mastercard said as of September 2025 that Start Path had brokered 15,000 connections and that participating startups had raised more than USD 25 billion after their time in the program.
What participants get, according to the program page, is access to Mastercard's network of partners and customers, opportunities to co-develop products for agentic environments, a community with demo days, and a dedicated fund that considers investment "if and when a strategic fit is identified". Eligibility requires a seed or later round already raised and a product live in market and generating revenue. However, whether any of the 22 companies received investment, and on what terms, has not been disclosed. Start Path is designed as an engagement program centered on network access rather than an accelerator built around investment, a point that matters for the assessment later in this article.
Reading the 22 companies in six layers
Looking at the 22 companies side by side shows which parts of agentic commerce Mastercard wants to leave to outside builders. The Paypers groups them into five clusters: payment infrastructure connections, risk and verification, commerce enablement, enterprise deployment, and vertical applications. Reorganized from a merchant's point of view, they form six layers: selling and checkout, catalog optimization, identity verification, fraud and audit, enterprise deployment, and vertical applications.
| Layer | Companies | Summary of official description | Relevance to merchants |
|---|---|---|---|
| Selling to agents and checkout | Firmly, Wizard, SolvaPay, Nevermined, Satsuma | Sell products through AI agents, accept agent payments, maintain LLM connectors | Most direct. This is where agent-driven revenue is booked |
| Catalog optimization | ReFiBuy | Continuously improve product catalogs so AI shopping agents can discover, recommend and buy them | Prerequisite for the selling layer. Product data quality decides exposure |
| Agent identity and verification | Skyfire, t54 | Give agents verified identity and payment credentials; verify them in real time on the merchant side | Shared basis for telling whose agent is transacting |
| Fraud, risk and audit | Tunic Pay, PayOS, Simetrik, Glassbox Labs | Contextual scam detection, intelligence on every party to a payment, transaction-level reconciliation, auditable workflows | Directly tied to chargebacks and reconciliation workload |
| Enterprise deployment | Grovity, 10Clouds, Sapiom, Merge | Business process automation, agent runtime for vertical AI, connecting business systems to AI models | System-side readiness for large retailers |
| Vertical and other | Luca AI, Thyris.AI, Masraff, MediConCen, Testudo, Crossmint | Conversational OS for SMBs, spend management, insurance claims, AI liability insurance, stablecoin infrastructure | Mostly outside e-commerce; limited direct relevance |
Two of these companies sit closest to merchant revenue and deserve a closer look.
Firmly is described officially as enabling merchants "to sell directly through any AI agents and digital channels with zero engineering effort, turning product discovery into instant checkout". Earlier in 2026 the company announced Firmly Connect, a single connection that covers multiple protocols including UCP, ACP, MCP, AP2 and TAP, and Best Buy's adoption has been reported. From the merchant's side this is an adapter layer that removes the need to guess which protocol will win. Mastercard's decision to include this layer reads as a search for outside distribution channels that carry Agent Pay to more merchants.
ReFiBuy is the only company in the cohort standing on the catalog side. Its official description is "an agentic commerce optimization platform that helps brands and retailers prepare and continuously improve product catalogs for discovery, recommendation and purchase by AI shopping agents". The company argues that AI agents evaluate product data directly and decide which products qualify, which get compared, and which are excluded before a shopper ever reaches a site. It also runs the AI Commerce Rankings with Digital Commerce 360, grading 1,000 retailers on AI shopping readiness. The fact that a product-data company, not a payments company, was admitted to a card network's program is important evidence that Mastercard recognizes transactions failing before they reach the payment step.
On the acceptance side, Nevermined is described as "a commerce logic layer between AI agents and money", letting merchants verify, accept and settle agent-driven purchases on the payment rails they already use. Its site promotes Card Delegation, which gives an agent a card and a budget cap, and states support for Visa virtual cards. Wizard is a consumer-side shopping agent that had already announced a three-way partnership with Mastercard and Stripe in May 2026, and now joins the cohort as well.
In the identity layer, Skyfire gives agents verified identity and payment credentials and promotes a Know Your Agent framework called KYAPay. t54 stands on the merchant and network side, supporting transactions with agents through real-time identity verification and risk assessment. The two occupy the sending and receiving ends of the same layer.
The eight companies in the fraud-and-audit and enterprise-deployment layers are tools for banks, payment providers and large retailers' system teams rather than vendors a typical merchant contracts directly. PayOS's claim to provide information on every entity in a payment, including the card, bank account, business, person, transaction and agent, before money moves is worth noting as an example of agents being formally counted among the parties to a payment. The six vertical companies include MediConCen in insurance claims and Testudo in AI liability insurance, which have little contact with e-commerce, and the "& Services" in the track name refers to this part.
How it fits with Agent Pay, Verifiable Intent and UCP
Over the past year and a half Mastercard has assembled its own set of components for agent payments. Agent Pay, announced in April 2025, uses Agentic Tokens issued per agent so the network can identify whose agent is paying and within what scope. Verifiable Intent, open-sourced with Google, records cryptographically what a user has delegated to an agent, and Agent Suite helps banks and merchants build agents.
Gaps remain. Agent Pay protects the moment of payment that the card network processes; it does not make a merchant's products discoverable to agents, nor does it let a merchant's systems receive an order that an agent sends. Mastercard joining UCP in January and working with Microsoft to put Agent Pay into Copilot Checkout are moves to cover that surface together with large platforms. This cohort fills in the rest with startups: ReFiBuy for catalogs, Firmly and Satsuma for merchant-side connections, Skyfire for agent-side identity, t54 for merchant-side verification, and Nevermined for payment logic, covering the inside of merchant systems and the agent-side implementations that a network itself finds hard to reach.
In the January announcement Mastercard said it builds standards and partnerships itself while also seeking "innovative newcomers", a description consistent with how the cohort is composed.
Visa, Stripe, and a caveat on the "ecosystem building" reading
Visa has chosen a different entry point in the same period. Visa Agentic Ready, formally launched in Europe in March 2026, is an environment where 21 banks and fintechs can test agent payments, an issuer-side program built around tokenization and biometrics. On the merchant side Visa offers the Trusted Agent Protocol to confirm an agent's legitimacy, and companies such as Crossmint have published agent card APIs built on Visa Intelligent Commerce. The contrast of Visa building its ecosystem from banks and Mastercard from startups holds, but neither company is leaving the other's territory open.
Stripe's Agentic Commerce Suite from April 2026 is the vertically integrated model, covering everything from merchant connections to Google Gemini integration. The room for Firmly and Nevermined lies in the gap between that vertically integrated model and the horizontal model of the card networks.
Forkast characterized the cohort as "ecosystem-building rather than protocol-building", arguing that Mastercard is using Start Path to validate and connect infrastructure builders rather than developing its own standard. Two caveats apply. First, Mastercard has in fact released Verifiable Intent and distributed Agent Pay specifications to issuers and merchants, so the premise that it is not building protocols does not match the record. Second, Start Path neither requires investment nor binds its participants. Nevermined advertises Visa virtual card support and Crossmint provides APIs for Visa Intelligent Commerce. Joining the cohort does not mean joining Mastercard's camp, and many of the 22 will keep operating across multiple networks. The lock-in effect is weaker than Forkast's framing suggests.
Which layer merchants should build first
What to take from the list of 22 is less the individual names than the fact that the merchant-side work of agentic commerce has reached the stage where it can be bought as products. Catalog optimization, cross-protocol connection, and acceptance of agent-initiated payments are each now offered by specialist vendors.
The catalog comes first. As ReFiBuy argues, agents make their selections on product data before a shopper reaches the site. Connecting checkout through Firmly does not keep a product in the running if its attributes, inventory and price cannot be read by agents. Conversely, once the catalog is in order, the connection layer can be swapped later regardless of which protocol wins.
Payment acceptance and identity verification come next. Once agent-driven orders start to grow, the decision is whether the payment provider in use supports Agentic Tokens and the Trusted Agent Protocol, and whether to insert a verification layer such as t54. Because network-side specifications in this layer are settling, choosing a PSP or adapter that already supports them is more rational than building in-house.
The fraud and audit layer can wait until agent-driven transactions become a non-negligible share of the total. The four companies in this cohort mainly serve banks and large enterprises, and the occasions on which a mid-sized merchant contracts them directly will be limited for now.
Conclusion
The 22-company first cohort of Start Path is structured so that outside startups cover the inside of merchant systems and the agent-side implementations that Mastercard's own Agent Pay and Verifiable Intent cannot reach. With Visa starting from banks and Stripe pursuing vertical integration, a card network bundling a layer of startups is a practical way to connect merchants without waiting for the standards battle to be decided.
The next things to watch are how many of the 22 ship products that actually connect to Agent Pay or UCP, and whether investments from the dedicated fund are made public. If the investments follow, Mastercard's commitment can be measured beyond the frame of the program.
