How to Connect SwitchWize to ChatGPT or Claude (MCP Setup Guide)

To connect SwitchWize to Claude or ChatGPT, add its free MCP server: paste one URL into Claude's connector settings, or run one npx command for a local client. No account or API key is required.

If you've asked an AI assistant something like "am I losing money keeping cash at my current bank," you've probably noticed it either declines to answer (it has no live data) or answers with a number it can't actually verify. Connecting an agent to a live data source fixes that. SwitchWize runs a free, read-only MCP (Model Context Protocol) server that exposes its rate, card, and index data as 15 callable tools, and setup is a copy-paste job, not a developer project.

What you get, in plain terms

Once connected, the agent you're talking to can call out to SwitchWize mid-conversation to pull live savings/CD/mortgage rates, estimate what a named bank's low rate is costing you on a given balance, check a credit card's current welcome offer or fee-refund window, look up what happens automatically when a CD matures, or run a full "what's my best move this week" pass across the cards you tell it you hold. It's a read-only connection: nothing it calls can move money, open an account, or change anything on SwitchWize's side or yours.

Option 1: Claude Desktop or Claude web (custom connector)

This is the path with no install step at all.

1. Open Claude's Settings, then Connectors.
2. Choose Add custom connector.
3. Paste the hosted endpoint: https://www.switchwize.com/api/mcp
4. Save. Claude will discover the 15 tools automatically the next time you start a conversation.

From there, just ask a normal question, like "compare the best current savings rates" or "what's my annual loss keeping $25,000 at Chase," and Claude will call the matching tool itself.

Option 2: A local client that only supports stdio (Claude Desktop's older config path, or similar tools)

Some clients expect a local process rather than a remote URL. For those, add this to the client's MCP server configuration:

{
  "mcpServers": {
    "switchwize": {
      "command": "npx",
      "args": ["-y", "switchwize-mcp"]
    }
  }
}

This runs switchwize-mcp via npx on demand. There's nothing to install ahead of time, and no separate process to manage.

Option 3: Cursor and other MCP-aware coding tools

Cursor, Windsurf, Cline, and similar tools read the same remote endpoint directly. Add a .cursor/mcp.json (or the equivalent file for your tool) with:

{
  "mcpServers": {
    "switchwize": {
      "url": "https://www.switchwize.com/api/mcp"
    }
  }
}

Option 4: ChatGPT and other apps, via the OpenAI Responses API

If you're building on top of an AI model rather than using a chat client directly, point the Responses API's mcp tool type at the same hosted endpoint:

const response = await client.responses.create({
  model: "gpt-5",
  tools: [{
    type: "mcp",
    server_label: "switchwize",
    server_url: "https://www.switchwize.com/api/mcp",
    require_approval: "never",
    allowed_tools: ["get_bank_gap", "get_top_rates"]
  }],
  input: "What could $25,000 at Chase be missing each year?"
})

Try it before connecting anything

You don't have to set any of this up to see what it returns. The developers page has a live playground: pick a sample question, click run, and see the real tool call and its real response, including the exact JSON an agent would receive.

What the answers actually look like

Every tool response carries more than a headline number. A typical response includes asOf/verified_at (when the underlying data was observed), a freshnessStatus (fresh, aging, stale, or unavailable, never silently omitted), and a methodologyUrl pointing at exactly how that figure is computed. See how AI agents should verify APY freshness for what to do with those fields once you have them, particularly the difference between a stale figure and a current one.

A note on what this isn't

This connects an agent to SwitchWize's own public rate and card data. It does not connect to your bank account, and it can't see your balances, transactions, or card numbers unless you type them into the conversation yourself for a calculation. If you're looking for the more advanced case of an AI agent actually reading (or moving money in) a real bank account, that's a different and much newer category of tool. See our comparison of the first bank-run MCP servers for how that's evolving separately.