Connect Your AI Agent to Slash
Slash now supports MCP (Model Context Protocol), letting you connect AI agents directly to your business finances. Create cards, send payments, manage invoices, and query your transaction data, all through Claude, Cursor, ChatGPT, or any MCP-compatible agent.¹
What you can do with an agent in Slash
Anything you'd normally do in the Slash dashboard, your agent can now do via prompt:
- Cards: Create virtual cards, set spend limits, lock to specific merchants
- Payments: Send ACH, wires, RTP, or crypto⁴
- Invoicing: Create and send invoices, view and approve reimbursements
- Crypto: Spin up deposit addresses to accept stablecoin payments from customers
- Analytics: Query your spend history and transaction data using SQL through our analytics endpoint
The key difference from other payment APIs is mandatory human approval. Agents can draft and submit actions, but execution only happens after your explicit sign-off. Because the agent never has unilateral control, prompt injection attacks cannot drain your Slash account.
How it works
1. Connect your agent
Add Slash's MCP server to your agent's configurator. In Cursor, that's ~/.cursor/mcp.json. In Claude Desktop or ChatGPT, add it through their MCP settings. Once connected, your agent has access to Slash tools, cards, payments, invoices, analytics, and more.
2. Approve actions
When your agent initiates a payment or creates a card, you'll get an approval request in Slack (or your configured channel). Review the details and approve or reject. The action only executes after you confirm.
3. Query your data
The analytics endpoint lets agents write SQL queries against your transaction history. Ask "what did we spend on software last quarter?" and your agent can pull the answer directly.
4. Security model
Agent-initiated payments are a security problem if done wrong. Our approach:
- Tokenized card details: When agents make purchases, they use RSA-encrypted tokens. The agent never sees raw card numbers or CVVs. Even if the agent is compromised via prompt injection, there's nothing to steal.
- Mandatory approval: Every financial action requires human confirmation. No exceptions.
- Scoped permissions: You control what the agent can access. Limit to read-only analytics, or enable full payment capabilities.
For companies building agents
If you're building agents that need to make payments on behalf of users, Slash can provide the infrastructure to get it done. Our MCP server handles card tokenization, approval flows, and transaction execution. You build the agent logic; we handle the money movement.
The shopify-mcp module (https://github.com/jason-slash/shopify-mcp) is open source and shows how to build E2E purchasing flows. It works with any Shopify store without API keys. To prove it works, we bought muffins with it.
Real-world examples
The Slash team has been using this internally to test it out. Here’s some examples of what we’ve been able to do with our in-house agent, Atlas:
- Automated Instacart orders: An agent can handle grocery ordering end-to-end, creating one-time-use cards with merchant locks and spend limits. Approvals come through Slack.
- Finance dashboard via prompt: Connect an AI and query your account to analyze spend patterns, investigate specific transactions, generate reports. Your finance dashboard is now a conversation.
- Submit and approve reimbursements: Settle expense reports by instructing your agent to reimburse employees for their purchases.
- Agentic invoice management: Agents can create, send, and track invoices. Review and approve reimbursement requests without logging into the dashboard.
- E2E Shopify checkout: Our open-source shopify-mcp module lets agents browse any Shopify store, add to cart, and complete checkout using Slash cards.
Slash for Agents is live right now. Connect your MCP client to get started.
Apply in less than 10 minutes today
Join the 5,000+ businesses already using Slash.
