Paystack MCP Server

A Model Context Protocol (MCP) server that enables AI assistants to interact with the full range of Paystack APIs.

Quick Start

Install and run via npm (recommended):

npx @paystack/mcp-server --api-key sk_test_your_key_here

Or for local development, clone and build:

git clone https://github.com/PaystackOSS/paystack-mcp-server.git
cd paystack-mcp-server
npm install
npm run build

Then configure your MCP client to use the server (see Client Integration).

Requirements

• Node.js v18+

• npm or yarn

• A Paystack test secret key (starts with sk_test_)

Configuration Options

You can provide your API key in two ways:

1. CLI argument (recommended): --api-key sk_test_...

2. Environment variable: Set PAYSTACK_TEST_SECRET_KEY

Security note: Only test keys (sk_test_*) are allowed. The server validates this at startup and will reject live keys.

Client Integration

The Paystack MCP Server works with any MCP-compatible client. Below is the standard configuration schema used by most clients (Claude Desktop, ChatGPT Desktop, Cursor, Windsurf, etc.).

Using npm (recommended)

For npm-installed server:

{
  "mcpServers": {
    "paystack": {
      "command": "npx",
      "args": ["@paystack/mcp-server", "--api-key", "sk_test_..."]
    }
  }
}

Using a local build

If you've cloned and built the server locally:

{
  "mcpServers": {
    "paystack": {
      "command": "node",
      "args": ["/path/to/paystack-mcp-server/build/index.js"],
      "env": {
        "PAYSTACK_TEST_SECRET_KEY": "sk_test_..."
      }
    }
  }
}

which node

where node

Where to add this configuration

How It Works

The Paystack MCP Server exposes the entire Paystack API to AI assistants by parsing Paystack's OpenAPI specification at runtime. Instead of hardcoding individual endpoints, the server dynamically discovers all available operations and makes them accessible through a small set of tools.

Available Tools

Available Resources

Example

When you ask your AI assistant something like "Get me the last 5 transactions on my Paystack integration", here's what happens behind the scenes:

1. The assistant calls get_paystack_operation("transaction_list") to look up the endpoint details

2. It gets back the method (GET), path (/transaction), and available query parameters

3. It then calls make_paystack_request with { method: "GET", path: "/transaction", data: { perPage: 5 } }

4. You get your transactions

Prompt recommendation

To get the best results when using this MCP server, be specific in your prompts and always include "Paystack" in your requests. The server provides built-in instructions and a knowledge resource (paystack://skill) that help the AI assistant find the right documentation, code snippets, and API details.

Good prompts:

• "Initialize a Paystack transaction for 50000 NGN"

• "Create a customer with email user@example.com on my Paystack account"

• "How can I send money with the Paystack API?"

• "Show me a cURL example for verifying a Paystack transaction"

Less effective prompts:

• "List my transactions" (unclear which service to use)

• "Charge a customer" (missing context about Paystack)

Being explicit ensures the LLM narrows down to the right tool quickly and reduces ambiguity.

Development

Run locally (without building)

For local development and testing, you can run the TypeScript source directly:

PAYSTACK_TEST_SECRET_KEY=sk_test_... npm run dev

Run with MCP Inspector

npm run inspect

Build

npm run build

Run tests

npm test

Troubleshooting

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

See CONTRIBUTING.md for more details.

License

MIT