What happened
Crossmint’s 18 November 2026 announcement describes Menta as an open-source agentic commerce plugin for ChatGPT, Devin, Grok Bot, Claude, Muse and other assistants. The company says Menta is listed in the ChatGPT, Devin and Grok Bot plugin marketplaces. For Claude, Muse and other assistants, users can copy a setup prompt from menta.crossmint.com into the assistant. Crossmint presents Menta as a consumer-facing layer built on its Agent Commerce Toolkit (ACT), which it says it launched in October as a set of APIs for a purchase flow from saving a card through checkout.
The workflow begins with a user connecting an account and adding a card through a secure page. The card is kept in a PCI-compliant vault, and the assistant does not see its number. Users can approve a budget for a defined period so purchases within that limit do not require repeated prompts, or require approval before every purchase. Crossmint says the assistant cannot approve payments or change cards. It can also use a card saved in a merchant account, Shop Pay, or local payment options such as Pix and Bizum. The company says Menta accepts cards issued worldwide.
For each store, Menta selects what Crossmint describes as the fastest available checkout route: the store’s own API, an agent protocol, or a browser built for checkout. A user can give the assistant product links and purchase details, such as size and delivery address. If a store requires sign-in, Crossmint says the user enters the password on a secure page rather than in chat, and Menta stays signed in for later use. The announcement says the system can handle purchases from requests such as ordering multiple items, but also warns that some store checkouts require steps an agent cannot complete.
Crossmint reports testing Menta at 3,000 stores, where it completed 91% of purchases and was 5–10 times faster than general-purpose agents using browser checkouts. The announcement does not describe the test protocol. For developers, Crossmint says Menta can be forked and shipped as an app, messaging bot or MCP server. Its card-capture component can be embedded in a developer’s page so the card number does not pass through that developer’s servers. Builders seeking more control can use ACT APIs, which Crossmint says provide the card vault, spending limits and checkout routing in one integration.
What the archive adds
New Runtime’s earlier coverage described Shopify’s Agentic Storefronts as infrastructure for selling through AI chats, and Anthropic’s commerce agents as shopper-facing workflows. Menta is a new event: Crossmint presents a reusable purchase plugin with card storage, spending controls and checkout routing rather than another store-facing surface. This connects to New Runtime’s earlier note on Shopify Agentic Storefronts only at the broader commerce-with-agents level; the actual linked note concerns Meta Muse availability, so it is not a direct Menta precedent.
Trend signal
This is a concrete example of agent-ready software exposing capabilities: Menta gives assistants bounded access to payment and checkout actions, while the user retains budget and approval controls. It also raises the operational question behind agent security moving to runtime boundaries, since card custody and spending authorization are enforced outside the prompt. Crossmint’s announcement describes the controls, but does not independently demonstrate their behavior under adversarial or failure conditions.
Why it matters
An agent that can recommend an item but cannot complete a transaction still leaves the most consequential step to a person. Menta’s design treats purchasing as a workflow with several distinct controls: a protected credential store, a budget that can persist over time, optional approval at each purchase, and a checkout route chosen for the merchant. That combination is potentially useful to builders who need agents to take actions with real financial consequences, not just retrieve information.
The integration boundary matters too. A team can build on Menta as an open-source product, embed its card-capture component, or use ACT for more direct control. These choices shift where a team must own the user experience, payment policy and failure handling. Crossmint’s claim of broad store coverage should not be read as universal reliability; the company itself says some checkouts cannot yet be completed. Builders should evaluate the user-facing recovery path as carefully as the successful purchase path.
Limits and uncertainty
The 91% completion rate across 3,000 stores and the 5–10× speed comparison are Crossmint’s own claims. The announcement does not specify the sample composition, purchase types, comparison-agent setup, test dates, or how a completed purchase was verified, and the results have not been independently checked here. Store checkout flows change, and Crossmint explicitly says Menta will not succeed at every store. The announcement describes card vaulting and user-defined limits but does not provide detailed technical documentation on enforcement, dispute handling, refunds, or the geographic scope of every payment method. It also does not quantify operating costs or disclose pricing for developers.
Skill to build now
Map one agent purchase flow as a sequence of permissioned steps: credential capture, budget approval, merchant selection, checkout, and receipt confirmation. For each step, define what the agent may do without interruption and what requires a person. Test both a low-risk purchase under a fixed budget and a checkout that requires sign-in; ensure credentials stay out of chat and give users a clear recovery route when automation fails.
Reader takeaway
Menta’s notable move is to combine payment credentials, user-set spending boundaries and store-specific checkout routing behind an assistant-facing plugin. That makes purchasing a more complete agent action, but not a guaranteed one: Crossmint’s own announcement reports a 91% test completion rate and acknowledges that some checkout flows remain out of reach.