How MCP Enables AI Purchasing

What is MCP?

Model Context Protocol (MCP) is an open standard created by Anthropic that defines how AI models connect to external tools and data sources. Think of it as a universal adapter: any MCP-compatible AI can use any MCP server, without custom integration for each tool.

MCP servers expose "tools" that the AI can invoke. A weather MCP server might expose a get_forecast tool. A calendar server might expose create_event. And a payment server might expose request_spend.

From tools to transactions

Most MCP tools are read-only or low-stakes: fetching data, searching files, checking the weather. But MCP doesn't limit what tools can do. If you give an AI access to a tool that can spend money, the AI can spend money.

This is where things get interesting — and where spend controls become essential.

A simple example

Imagine you're chatting with Claude and you say:

"Find me a highly-rated portable charger under $40 and order it."

Without MCP payment tools, Claude can research options and recommend one. But it can't complete the purchase — you'd need to click through checkout yourself.

With an MCP payment tool like PettyBot, Claude can:

1. Research portable chargers and find the best option
2. Call request_spend with the merchant, amount, and item details
3. Wait for your approval (you get a push notification)
4. Complete the purchase once approved

The AI handles the legwork. You stay in control of the money.

Why approval matters

Giving AI agents spending ability without guardrails would be reckless. Even well-intentioned AI can misunderstand requests, find the wrong product, or make mistakes.

That's why the approval layer is non-negotiable:

• Every purchase requires explicit approval — You see the merchant, amount, and description before any money moves.
• Spend limits act as guardrails — Even if you accidentally approve, the transaction can't exceed your configured limits.
• Merchant controls add precision — Allow specific merchants, block categories, set per-merchant limits.

The AI proposes. You approve. That's the model.

What makes a good AI purchase?

Not every purchase is a good fit for AI agents. The sweet spot is research-intensive, one-off purchases where the AI's ability to compare options adds real value:

• Specialty items — "Find a 1000/6000 grit whetstone for my Japanese knives"
• Gift research — "Find a birthday gift for my nephew who's into dinosaurs, under $30"
• Travel booking — "Book the cheapest direct flight to Denver next Friday"
• Niche retailers — Purchases from specialty stores where you don't have an account

These aren't your Amazon reorders or DoorDash lunches. Those merchants already have your card on file. AI purchasing shines when there's research to do and a merchant to discover.

The technical flow

For developers curious about the implementation:

1. MCP connection — Your AI client (Claude Desktop, custom agent, etc.) connects to the PettyBot MCP server.
2. Tool discovery — The AI discovers available tools: request_spend, check_balance, list_transactions.
3. Spend request — When the AI decides to purchase, it calls request_spend with merchant, amount, description, and item details.
4. User approval — PettyBot sends a push notification. The user sees a card with all details and approves or denies.
5. Payment execution — On approval, PettyBot issues a scoped payment token and completes the transaction via Stripe.
6. Confirmation — The AI receives confirmation and can report success to the user.

Where this is headed

MCP adoption is accelerating. As more AI assistants support the protocol, payment tools like PettyBot become universally accessible. The same approval flow works whether you're using Claude, a custom agent, or whatever comes next.

The future isn't AI spending without oversight. It's AI handling the tedious parts of commerce — research, comparison, checkout — while you retain final say on every dollar spent.

Try it yourself

PettyBot is currently in private beta. Join the waitlist to get early access when we launch.

• PettyBot home
• Frequently asked questions
• What is the Agent Commerce Protocol?
• Glossary of terms