CardPilot Remote MCP Server
This is a remote MCP (Model Context Protocol) server for CardPilot. It exposes tools to fetch and analyze credit card data.
Server URL
Public URL: https://mcp.cardpilot.ca/sse
Note: This server uses a custom domain to ensure compatibility with OpenAI Agent Builder.
Available Tools
1. get-cards
Fetches a list of credit cards with detailed metadata, suitable for ranking and comparison.
Input parameters:
| Parameter | Type | Description |
|---|---|---|
sort | string (enum) | Sort criteria: recommended, welcome_offer, interest_rate, annual_fee, net_value |
direction | string (enum) | Sort direction: asc, desc |
ids | string | Comma-separated list of card IDs (e.g., card-a,card-b) |
bank | string | Filter by bank name (e.g., "TD", "RBC") |
category | string | Filter by category (e.g., "travel", "cash back") |
noFee | boolean | Set to true to filter for no-annual-fee cards |
limit | number | Maximum number of cards to return (default: 5) |
persona | string (enum) | Target persona for tailored ranking: average, student, newcomer, premium |
Output structure:
{
"cards": [
{
"cardId": "string",
"name": "string",
"bank": "string",
"annualFee": number,
"score": number,
"details": { "rewards": number, "perks": number, "fees": number, ... }
}
],
"meta": { "total": number, "sort": "string", "direction": "string" }
}
2. get-guides
Fetches a list of educational guide metadata. This tool is optimized for chatbots to provide links to full articles.
Input parameters: None
Output structure:
{
"guides": [
{
"slug": "string",
"title": "string",
"intro": "string",
"icon": "string (emoji)",
"url": "string"
}
]
}
Agent System Prompt
If you are using this MCP server with an LLM Agent (like OpenAI Custom GPT), add the following to your System Instructions:
You are an expert credit card advisor powered by CardPilot data.
1. **Card Recommendations**:
- ALWAYS use the `get-cards` tool to fetch real-time data before making recommendations.
- Set `limit=5` by default to ensure concise recommendations.
- If the user specifies or asks for a specific persona (e.g., student, newcomer), set the `persona` parameter (e.g., `persona="student"`).
- Use the `sort` parameter to align with user priorities (e.g., `sort="annual_fee"` for cheap cards).
- Use filters like `bank="TD"` or `category="travel"` to narrow down results.
- For "no fee" requests, explicitly set `noFee=true`.
- When presenting cards, list the Name, Annual Fee, Welcome Bonus, and a brief "Why it fits" explanation.
- Link the card name to the `applyUrl` or CardPilot detail page if available.
2. **Educational Content**:
- If a user asks general questions (e.g., "Cash back vs Points"), use `get-guides` to see if there is a relevant article.
- Provide the answer based on the guide's `intro` and encourage the user to read the full guide by providing the `url`.
3. **General Rules**:
- Do not make up card details. If data is missing, state that.
- Be concise and helpful.
How to Use
OpenAI Agent Builder
- Create a new Agent.
- Under Actions, click Add Action.
- Select "Add from URL".
- Enter:
https://mcp.cardpilot.ca/sse - It should load the
get-cardstool immediately.
Cloudflare AI Playground
- Go to https://playground.ai.cloudflare.com/
- Enter the server URL:
https://mcp.cardpilot.ca/sse - Click Connect.
Claude Desktop
Add the following to your claude_desktop_config.json:
{
"mcpServers": {
"cardpilot": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://mcp.cardpilot.ca/sse"
]
}
}
}
Troubleshooting
OpenAI 424 Error ("Unable to load tools")
If you see this error, ensure you are using the Custom Domain URL (mcp.cardpilot.ca).
OpenAI blocks generic .workers.dev domains for MCP servers.
Local Development
To test locally:
- Install dependencies:
npm install
- Start local server:
npm start
(Runs on port 8787)
- Expose via Ngrok (Required for OpenAI testing):
ngrok http 8787
Use the Ngrok URL (e.g., https://xxxx.ngrok-free.app/sse) in Agent Builder.
Deployment
Deploy to Cloudflare Workers:
npm run deploy
The server will be live at https://mcp.cardpilot.ca/sse.