Virtual cards for AI agents.
Give your agent a card it can spend safely. One-time virtual cards with spend caps and human approvals.
Agentcard gives AI agents a safe way to pay for things. Users and companies fund a wallet, set spend caps and approval rules, and agents mint one-time virtual cards at checkout. Each card works once, at any merchant that accepts cards, so a leaked number is worthless and spend never exceeds the cap. Agents connect through a remote MCP server (OAuth 2.0 with dynamic client registration), a REST API with OAuth client credentials, or the agent-cards CLI. Built for agent developers: checkout detection, card autofill, transaction feeds, and human-in-the-loop approvals are all first-class tools.
| URL | https://mcp.agentcard.sh/mcp |
| Transport | Remote, streamable HTTP |
| Auth | OAuth 2.0, automatic. Dynamic client registration + PKCE (S256). No API keys to paste. |
On first connect, your MCP client registers itself and opens a browser for you to sign in. There are no keys to copy. The same URL serves both test and live accounts; which one you get is decided at sign-in, never by the address.
Pick your client. Every config points at the same URL and uses the same automatic OAuth flow.
claude mcp add --transport http agentcard https://mcp.agentcard.sh/mcpThe first time your agent calls a tool, Claude Code opens a browser to authorize. You can also run npx agent-cards@latest setup-mcp, which signs you in and wires the server up for you.
Open Settings → Connectors → Add custom connector, then paste:
https://mcp.agentcard.sh/mcp
Claude walks you through sign-in on first use.
Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):
{
"mcpServers": {
"agentcard": {
"url": "https://mcp.agentcard.sh/mcp"
}
}
}Add to .vscode/mcp.json:
{
"servers": {
"agentcard": {
"type": "http",
"url": "https://mcp.agentcard.sh/mcp"
}
}
}Add to your cline_mcp_settings.json (see also llms-install.md, which Cline can follow on its own):
{
"mcpServers": {
"agentcard": {
"type": "streamableHttp",
"url": "https://mcp.agentcard.sh/mcp"
}
}
}Add to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"agentcard": {
"serverUrl": "https://mcp.agentcard.sh/mcp"
}
}
}Point it at the remote server. Any client that supports remote MCP servers with OAuth will work:
{
"mcpServers": {
"agentcard": {
"url": "https://mcp.agentcard.sh/mcp"
}
}
}A selection of what the server exposes. Descriptions are shortened; the live server carries the full set.
| Tool | What it does |
|---|---|
create_card |
Mint a virtual card, funded from the wallet. |
get_card_details |
Return card number, CVC, and expiry to fill a payment form. May require human approval. |
fill_card |
Fill an existing card into the current checkout form via the Agentcard Pay browser extension. |
close_card |
Permanently close a card. Irreversible, and safe to call twice. |
fund_wallet |
Add funds to the wallet with Apple Pay or Google Pay. |
check_balance |
Check a card's live available balance without exposing its credentials. |
get_wallet |
Show the funding wallet and its balance. |
list_transactions |
List a card's transactions: amount, merchant, status, timestamps. |
detect_checkout |
Detect whether the current browser tab is a checkout page. |
pay_checkout |
Create or reuse a card and auto-fill the checkout form. |
buy |
Shop and check out in natural language across linked merchants. |
approve_request |
Resolve a pending human-approval request (approve or deny). |
start_kyc |
Run conversational identity verification when a card requires it. |
whoami |
Show the authenticated account: email, plan, and status. |
Once connected, ask your agent to call whoami to confirm the connection, then get_wallet to see the funding wallet.
- The agent connects and authorizes over OAuth (once).
- It funds a wallet, or a company pre-funds one, with a spend cap.
- At checkout the agent mints a fresh virtual card for that purchase.
- The card authorizes once, within the cap, then closes.
A stolen card number is worthless, and no single purchase can exceed the limit you set.
- Product: https://agentcard.sh
- Docs: https://docs.agentcard.sh
- Privacy: https://www.agentcard.sh/privacy
- Terms: https://www.agentcard.sh/terms
- Support: felipe@agentcard.sh
MIT. See LICENSE.
