Agent Wallet Simulator
Give AI agents a fake-money wallet with spend limits and approvals, then watch every transaction live. No real money.
Frequently Asked Questions
What is the agent wallet simulator?
It gives AI agents a wallet full of fake money (SIM-USD) and a real payments API to spend it through. Each agent has its own key, balance and spending policy, and every payment shows up in the ledger on this page the moment it happens. It is a safe way to see how an agent behaves with money before giving it a real card.
Does any real money move?
No. Balances are simulated, the currency is SIM-USD, and there is no way to deposit or withdraw real money. Merchants are just names, and the built-in shop hands out fake goods.
How do I connect my AI agent?
Copy the Markdown instructions on this page (with the agent key included) into your agent's system prompt, or download them as SKILL.md. They explain how to check the balance, pay with an Idempotency-Key, handle declines and approvals, and buy from the shop. The agent only needs to make HTTP requests, so any framework or a plain curl tool works.
Can I connect the wallet as an MCP server?
Yes. Each wallet is also a remote MCP server at its base URL plus /mcp, authenticated with the agent key. MCP clients such as Claude Code get tools like check_balance, get_quote, pay, get_receipt and redeem_item, with the same mandates, limits and approvals as the HTTP API. The page has a ready-made Claude Code command and JSON config.
What spending rules can I set?
For each agent: a per-payment limit, a rolling 24-hour limit, a threshold above which payments need your approval, and a freeze switch. Merchants are denied by default: you list the ones the agent may pay, each with optional per-merchant caps, or deliberately switch on Allow any merchant. Payments that break a rule are declined with a reason the agent can read, such as over_daily_limit or merchant_not_allowed.
What is a mandate?
A mandate is your signed authorization for one task: which agent, which merchants (and optionally which items or categories), how much per payment and in total, and until when. Every payment must name a mandate, and the server checks the signed terms before money moves. The agent's memo is just a note; the mandate is the proof. Each new agent gets a default testing mandate you can revoke.
How do approvals work?
A payment above the approval threshold is not declined; it waits with the money on hold, the way a card authorization does. You approve or reject it in the ledger, and the agent sees the result by polling the transaction. If nobody decides within 15 minutes, it expires and the hold is released.
What is the 402 shop?
A pretend store that sells resources the way paid APIs for agents do. Requesting an item without paying returns 402 Payment Required with a quote whose price is frozen for 5 minutes. The agent pays that quote, then repeats the request with the receipt to get the item. A receipt is bound to the agent that paid, single-use, and expires after 10 minutes. You can change shop prices to test how an agent handles a price move.
How long does a simulation last?
A simulation is deleted after 24 hours without activity and with no tab open watching it, or immediately if you delete it. The owner key and agent keys are stored only in this browser; the server keeps only their hashes. You can rotate or revoke any agent's key, and issue read-only keys.
Related Tools
Rate Limiter Simulator
Get a free URL backed by a real token bucket. Hit it with curl and watch it drain and refill live.
Webhook Tester
Get a free temporary webhook URL and inspect every request it receives, live: method, headers, query, and body.
MCP Config Generator
Generate Model Context Protocol (MCP) server configs for Claude Desktop, Claude Code, Cursor, VSCode, and Windsurf.
Token Counter
Estimate the token count for text with various LLM models.