Building with AI

Use AI coding assistants like Claude Code, Cursor, and Codex to explore, build, and debug with the Grid API

Grid’s documentation, OpenAPI spec, and CLI are designed to work with AI coding assistants like Claude Code, Cursor, and Codex. Whether you’re exploring the API for the first time or building a production integration, AI tools can help you move faster.

What AI assistants can do with Grid

AI-accessible documentation

These Grid docs are automatically available to LLMs and AI tools in machine-readable formats — no configuration needed.

llms.txt

Grid docs generate llms.txt files that give AI tools a structured index of all documentation:

• /llms.txt — Concise index of all pages with titles and descriptions
• /llms-full.txt — Complete documentation content in a single file

These are generated automatically and always up to date. Use llms-full.txt when you want to give an AI assistant full context about the Grid API in one shot.

Markdown export

Each page in the Grid docs is automatically available as a Markdown file simply by adding .md to the end of the URL.

For example, the Building with AI page is available as /platform-overview/building-with-ai.md.

You can also copy any page’s content as Markdown with the keyboard shortcut Command + C (Ctrl + C on Windows) and paste it directly into ChatGPT, Claude, or any AI assistant for context-aware help with your specific question.

Install the Grid API agent skill

The Grid API skill gives Claude Code or another Skill-compatible agent full access to the Grid API via a built-in CLI. Install it with:

npx skills add lightsparkdev/grid-api

Make sure to install it for whichever agent you’re using. For example, if you’re using Claude Code, you’ll need to explicitly select Claude Code in the agent installation selection screen.

Once installed, you can start asking questions immediately. To execute API operations, you’ll need to configure your credentials.

Configure your credentials

To set your grid credentials, simply ask the agent to help you configure them:

Help me configure my Grid API credentials.

It’ll prompt you for your API Token ID and Client Secret, validate them, and save to ~/.grid-credentials for future use.

Start in the sandbox environment to experiment safely. The Skill is great at generating fake account data to help you test different flows.

MCP server

Grid provides a Model Context Protocol (MCP) server that gives AI agents access to Grid’s documentation, so your agentic workflows have full context of the Grid API during implementation.

https://mcp.grid.lightspark.com/

The MCP server exposes two tools:

• search_grid_api_documentation — Search the Grid docs for relevant guides, code examples, and API references
• get_page_grid_api_documentation — Retrieve the full content of a specific documentation page

The MCP server provides documentation access to support agentic implementation workflows. It does not yet support interacting with the Grid API directly — use the Grid API agent skill or direct API calls for that.

Create an API token

The MCP server requires Grid API credentials. In the Grid dashboard, create a read-only scoped API token and copy the client ID and client secret. You’ll plug these into the x-grid-client-id and x-grid-client-secret headers in the configurations below.

Configure your client

Claude Code

In Claude Code, ask Claude to update your .claude.json config with:

{
  "mcpServers": {
    "grid_mcp": {
      "type": "http",
      "url": "https://mcp.grid.lightspark.com/",
      "headers": {
        "x-grid-client-id": "{client id}",
        "x-grid-client-secret": "{client secret}"
      }
    }
  }
}

Claude desktop app

In the Claude desktop app, open Settings → Developer, edit claude_desktop_config.json, and add:

{
  "mcpServers": {
    "grid_mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.grid.lightspark.com/",
        "--header",
        "x-grid-client-id: {client id}",
        "--header",
        "x-grid-client-secret: {client secret}"
      ]
    }
  }
}

The desktop app uses mcp-remote to bridge to a remote HTTP MCP server.

Cursor

In Cursor, add the server to ~/.cursor/mcp.json (global) or .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "grid_mcp": {
      "url": "https://mcp.grid.lightspark.com/",
      "headers": {
        "x-grid-client-id": "{client id}",
        "x-grid-client-secret": "{client secret}"
      }
    }
  }
}

ChatGPT and others

Any client that supports MCP’s streamable HTTP transport (ChatGPT, Windsurf, Zed, Cline, and others) can connect using the same URL and headers:

• URL: https://mcp.grid.lightspark.com/
• Headers: x-grid-client-id: {client id} and x-grid-client-secret: {client secret}

For clients that don’t yet support HTTP-transport MCP servers natively, use mcp-remote as a bridge — see the Claude desktop app tab for an example.

Example prompts

Try these prompts in Claude Code or paste them into your AI assistant of choice:

Getting started

What currencies does Grid support? Show me the coverage by country.

Walk me through the steps to send a payout from USD to MXN via CLABE.

Payouts

Create an external account for a bank in Mexico using CLABE, then create
a quote to send $500 USD and show me the exchange rate before executing.

Send $100 to $[e-mail] via UMA. Show me the exchange rate first.

On/off-ramps

Help me set up a fiat-to-crypto on-ramp. I want to convert USD to BTC
using a Spark wallet.

I need to off-ramp USDC to a US bank account. Walk me through the full flow.

Account management

List all my customers and their KYC status in a table.

Create a new individual customer and generate a KYC onboarding link.

Debugging

I'm getting QUOTE_EXPIRED errors. What's happening and how do I fix it?

My external account creation is failing for a Nigerian bank account.
What fields am I missing?

Tips for best results

1. Be specific about your use case — “Send a payout to Brazil via PIX” gets better results than “help me send money”
2. Start with the sandbox — Ask the AI to use sandbox mode so you can experiment without real funds
3. Give context — Paste the relevant docs page or point the AI to /llms-full.txt for full API context
4. Iterate on errors — If an API call fails, paste the error and ask the AI to diagnose it