This is a TypeScript example project demonstrating how to connect to and use the Banked MCP (Model Context Protocol) server. This example uses Mastra as a framework, but the focus is on showcasing Banked's MCP server functionality.

The Banked MCP server provides payment-related tools and capabilities that can be integrated into AI agents and workflows. This example shows how to:

• Connect to the Banked MCP server
• Authenticate using OAuth
• Use Banked payment tools in your agents
• Filter available tools for specific use cases

• Node.js (v18 or higher)
• npm

1. Clone or download this repository
2. Install dependencies:

   npm install

1. Create a .env file by copying .env.example:

   cp .env.example .env

2. Edit the .env file and set up the following required environment variables:

   Banked OAuth Credentials (Required):

   • BANKED_OAUTH_CLIENT_ID - Your Banked OAuth client ID
   • BANKED_OAUTH_CLIENT_SECRET - Your Banked OAuth client secret
   • BANKED_OAUTH_CLIENT_SCOPE - OAuth scope should be your business application ID

   Important: To obtain Banked OAuth credentials, please contact [email protected]

   Model Provider Configuration (Required):

   Set MODEL_PROVIDER to select which provider to use:

   • MODEL_PROVIDER=google-vertex - Use Google Vertex AI
   • MODEL_PROVIDER=anthropic - Use Anthropic
   • MODEL_PROVIDER=openai - Use OpenAI

   Then configure the required credentials for your selected provider:

   For Google Vertex AI (MODEL_PROVIDER=google-vertex):

   • GOOGLE_CLOUD_PROJECT - Your Google Cloud project ID (required)
   • GOOGLE_CLOUD_LOCATION - Google Cloud location (optional, default: europe-west1)
   • MODEL_NAME - AI model name (optional, e.g., gemini-2.5-flash)

   For Anthropic (MODEL_PROVIDER=anthropic):

   • ANTHROPIC_API_KEY - Your Anthropic API key (required)
   • MODEL_NAME - AI model name (optional, default: claude-sonnet-4-5)

   For OpenAI (MODEL_PROVIDER=openai):

   • OPENAI_API_KEY - Your OpenAI API key (required)
   • MODEL_NAME - AI model name (optional, default: gpt-4o)

   Tip: See .env.example for a complete example configuration file with all three provider options documented.

   Optional Configuration:

   • BANKED_MCP_URL - Banked MCP server URL (default: https://mcp.banked.com)
   • BANKED_OAUTH_URL - OAuth token endpoint (default: https://api.banked.com/oauth/token)
   • BANKED_ALLOWED_TOOLS - Comma-separated list of allowed tools (for filtering)

Start the development server:

npm run dev

Once the server is running, open your browser and navigate to:

http://localhost:4111

This will open the interface where you can interact with the example agents that use the Banked MCP server.

Once connected, you can interact with Banked's tools through natural conversation. Here are some examples:

Try asking:

Create a $50 payment for invoice INV-2024-001

The assistant will use payments.create and return a checkout URL. You can also ask about the tool's parameters:

What parameters does payments.create need?

You can search for payments in several ways:

Find the payment for invoice INV-123

or

Show me payments from last week

The assistant will use payments.search with the appropriate filters.

To process a refund, you can specify the payment directly:

Refund $25 from payment [payment-id]

or reference it more naturally:

Refund $25 from last payment

The assistant will use payments.refund to process the refund.

This project includes two example agents that demonstrate different ways to use the Banked MCP server:

• Basic Agent - Connects to Banked MCP server with full access to all available tools
• With Tool Filter Agent - Connects to Banked MCP server with filtered tool access (using BANKED_ALLOWED_TOOLS)

Both agents are configured to help users with payment-related questions and can interact with Banked's payment services through the MCP server.

• For Banked OAuth credentials and MCP server support: [email protected]