We just shipped something we're excited about: an MCP server that lets you build AI agents with real payment capabilities — directly from your coding environment.

No context switching. No Postman. No reading API docs for an hour. Just connect, ask your AI, and go.

What is MCP?

MCP (Model Context Protocol) is a standard created by Anthropic that lets LLMs connect to external tools. Think of it as USB for AI — plug in a server, and your AI assistant gets new capabilities.

When you connect the Shatale MCP server, your AI can:

Try It Now (No Signup Required)

``bash

npx shatale-mcp-server

`

That's it. The server starts in guest mode with 4 tools available immediately:

  • shatale_list_mcc_codes — Browse merchant category codes for policy design
  • shatale_demo_scenarios — See 5 pre-built scenarios showing the policy engine in action
  • shatale_generate_snippet — Get ready-to-use TypeScript or curl code
  • shatale_get_started — Setup instructions and signup link

No API key, no account, no credit card.

Full Sandbox Mode

Sign up at [admin.shatale.com/register](https://admin.shatale.com/register?ref=blog-mcp) (free, takes 30 seconds), grab your sandbox key, and unlock 5 more tools:

  • shatale_create_demo_agent — Create an agent with card and policy in one step
  • shatale_simulate_payment — Run a transaction simulation with full rule traces
  • shatale_explain_decision — Get a detailed breakdown of why a payment was approved or declined
  • shatale_test_scenarios — Batch-test 4 different transactions against your setup
  • shatale_get_agent_status — Check agent health and recent transaction history

How It Looks in Practice

Here's a real conversation with Claude Desktop:

You: Create a shopping agent with a €1,000 monthly budget. Block gambling and alcohol.

Claude: (calls shatale_create_demo_agent) Done. Agent created with:

  • Virtual card issued
  • Policy: €500 per-transaction limit, €1,000/month, MCC 7995 (gambling) and 5921 (alcohol) blocked

You: Simulate buying sneakers for €150 at Nike Store.

Claude: (calls shatale_simulate_payment) APPROVED

  • spend_limit: passed — €150 within €500 per-tx limit
  • mcc_check: passed — MCC 5691 (Clothing) is allowed
  • balance_check: passed — Balance €1,000 covers €150

You: Now try €50 at an online casino.

Claude: (calls shatale_simulate_payment) DECLINED

  • spend_limit: passed — €50 within limit
  • mcc_check: failed — MCC 7995 (Gambling) is blocked by policy

The policy engine works exactly as configured. No surprises.

Supported Editors

We support every major MCP-compatible editor:

  • Claude Desktop
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • Zed
  • Continue

Copy-paste configs for each editor are on our [MCP docs page](https://shatale.com/mcp).

What You Can Build

The MCP server is designed for developers building AI agents that need to make purchases:

Shopping agents — browse, compare, and buy products within a budget. Block luxury, gambling, or any category you want.

Travel agents — book flights and hotels. Allow only airlines, hotels, and car rentals.

Procurement agents — company purchasing within approved vendor categories and department-level budgets.

Expense management — employee agents with daily limits, no alcohol/tobacco, and real-time approval workflows.

Security by Design

A few things we made sure of:

  • Sandbox only — Production API keys are rejected. You can't accidentally use real money through MCP.
  • No card data — PAN, CVV, and card numbers are never exposed through MCP tools.
  • Local transport — The server runs as a local stdio process. No network ports, no remote connections.
  • Open source — Full source on [GitHub](https://github.com/solskiysb/shatale-mcp-server). MIT license.

What's Next

This is v0.1 — the foundation. We're planning:

  • More policy templates for common verticals
  • Webhook monitoring from your IDE
  • Multi-agent setups with shared budgets
  • Listings in MCP registries (Smithery, Glama)

Get Started

`bash

npx shatale-mcp-server

``

Or visit [shatale.com/mcp](https://shatale.com/mcp) for setup guides, tool reference, and use case examples.

Questions? Open an issue on [GitHub](https://github.com/solskiysb/shatale-mcp-server) or reach us at dev@shatale.com.