We get asked this constantly: "Does AgentCard work with [my agent]?" Short answer — probably yes. But the details matter. Some agents can browse the web and fill out forms. Some can call APIs but can't touch checkout pages. Here's what we've seen across the landscape in 2026.
What an agent needs to make a purchase
To buy something online, an agent needs two capabilities:
- Web interaction — the ability to navigate to a website, fill in form fields (card number, CVV, expiry, billing address), and click buttons
- Card details — a valid payment method it can enter at checkout
The first part depends on the agent. The second part is what we solve — give the agent a prepaid virtual Visa, and it has real payment details to use at checkout.
Agent-by-agent breakdown
Claude Code
Can make purchases: Yes
Claude Code is honestly our most popular agent right now. It executes shell commands, makes HTTP requests, and interacts with web services. With our CLI installed, Claude Code directly runs agentcard details <id> to retrieve card details and uses them programmatically. It also reads our skill endpoint for structured instructions — so it basically figures out AgentCard on its own.
Cursor
Can make purchases: Yes
Cursor has terminal access and runs CLI commands natively. Retrieve AgentCard details, complete purchases — it handles the whole flow. Cursor also supports MCP, which makes the integration even smoother.
Devin
Can make purchases: Yes
Devin has a full browser environment. It navigates to checkout pages, fills in card details, clicks "buy." One of the most capable agents for end-to-end purchasing — we've been impressed watching it work.
OpenAI Codex
Can make purchases: Yes (with setup)
Codex executes code and makes API calls well. It can grab card details from our CLI and use them in automated checkout flows. Browser-based checkout needs extra setup, though — Codex isn't as strong there.
GitHub Copilot
Can make purchases: Limited
Copilot mostly lives inside code editors. Limited web interaction. It can write code that uses card details programmatically — calling an API that requires payment, for instance — but don't expect it to fill in a checkout form.
Windsurf
Can make purchases: Yes
Windsurf has terminal access and can run shell commands, making it compatible with the AgentCard CLI workflow. It can retrieve card details and use them in automated flows.
ChatGPT (with browsing)
Can make purchases: Limited
ChatGPT can browse the web and describe checkout flows, but its ability to directly fill in payment forms is limited by its browsing sandbox. It works best when card details are used programmatically rather than through form-filling.
Custom agents
Can make purchases: Yes
If you're building your own agent, you have full control. Use the AgentCard CLI to create and retrieve cards, then integrate the card details into your agent's checkout logic.
The MCP advantage
This is worth calling out separately. Agents that support MCP — like Claude Code and Cursor — have a real edge here. MCP lets agents discover and use tools in a structured, standardized way.
What this means in practice: the agent doesn't need to be taught how to use AgentCard. It reads the skill definition at agentcard.ai/skill and just... figures it out. We've watched Claude Code go from zero knowledge to completing a purchase in under a minute, without any prompting about how AgentCard works.
What about security?
Look, we know this makes people nervous. Fair enough. Here are the questions we hear most:
- What if the agent gets prompt-injected? — With a prepaid card, the damage is limited to the card balance. Load $25 and that's the maximum exposure.
- What if the agent makes repeated purchases? — The prepaid balance acts as a hard cap. Once depleted, all further charges are declined.
- What if the card details leak? — Freeze the card instantly from the CLI. Since it's prepaid with a small balance, the exposure is minimal.
Getting started
No matter which agent you're running, setup is the same:
- Install AgentCard:
npm install -g agentcard - Sign up:
agentcard signup --email you@example.com - Buy a card:
agentcard create --amount 25 - Get details:
agentcard details <card-id> - Hand the details to your agent
Step 5 is where agents differ. CLI-based agents (Claude Code, Cursor, Windsurf) run the entire flow autonomously — agentcard create to agentcard details to checkout — without you lifting a finger. Devin does the same through its terminal. Custom agents can call the CLI directly or integrate programmatically. The point is: the payment part is solved. The rest depends on your agent's capabilities.