MCP for ChatGPT

ChatGPT can use CurrencyCore as a custom MCP server on the web. You add the server in the workspace plugins or connectors area, connect it with OAuth, and then call the currency tools from a chat.

Checked 2026-11-10 against OpenAI’s connector documentation. Menu labels vary by plan and rollout, so the exact wording in your workspace may differ.

Server URL:

https://mcp.currency-core.com/mcp

Plan and workspace requirements

If you do not see a custom server option, the cause is usually the plan or a workspace policy, not the CurrencyCore URL. Check the current OpenAI documentation for your workspace type rather than relying on a plan name.

Setup

1. Confirm you are in the right workspace. Custom MCP servers are a workspace setting, not a personal preference.
2. Open the workspace plugins or connectors area. This is in workspace settings. On some plans it appears as a plugins directory with an option to add a custom MCP server.
3. Choose the option to add a custom MCP server.
4. Enter a name and the server URL. Use CurrencyCore for the name and https://mcp.currency-core.com/mcp for the URL.
5. Connect the server. ChatGPT starts the OAuth flow, which normally opens a browser for sign-in.
6. Sign in to CurrencyCore and choose an organization. The connection is bound to that organization.
7. Start a new chat and send a test prompt.

There is no JSON config file for this path. If a guide tells you to paste mcpServers JSON into ChatGPT, it is describing a different client.

Verify

Ask:

Use the CurrencyCore convert tool to convert 500 USD to INR and include the rate date.

A working connection returns a conversion and a date. If ChatGPT answers from memory, confirm the server is connected for that workspace and that the tool is enabled in the chat.

What a successful call returns

A successful tool call returns the REST body as text, a structured copy for clients that support it, and metering metadata:

{
  "content": [
    {
      "type": "text",
      "text": "{\"from\":\"USD\",\"amount\":500,\"date\":\"2026-11-10\",\"results\":[{\"to\":\"INR\",\"result\":41725.0,\"rate\":83.45}]}"
    }
  ],
  "structuredContent": {
    "from": "USD",
    "amount": 500,
    "date": "2026-11-10",
    "results": [{ "to": "INR", "result": 41725.0, "rate": 83.45 }]
  },
  "_meta": {
    "currencycore/tool_calls": 1,
    "currencycore/billed_requests": 1,
    "currencycore/status": 200
  }
}

Read structuredContent when your client supports it. Read content[0].text when it does not. Check _meta["currencycore/billed_requests"] to confirm the call counted against the organization; a public reference call or a failed call returns 0.

Tool refresh behavior

ChatGPT freezes the tool list it discovered when the connector was added. If the CurrencyCore server adds or changes tools, an existing chat may not see the change.

• Start a new chat after a server change.
• Refresh or reconnect the connector if the option is available.
• If a tool you expect is missing, remove and re-add the connector, then start a new chat.

The 14 CurrencyCore tools and their plans are listed in MCP tools. Client-side caching can make a newly added tool appear late.

What ChatGPT can call

Core tools: convert, get_rates, get_rates_by_base, get_history, get_ppp_factor, plus the public list_countries and list_currencies.

Analytics tools: analyze_history, analyze_ppp, analyze_volatility, analyze_correlation, analyze_drawdown, analyze_safe_haven, analyze_mean_reversion. These require Growth or higher.

Each successful keyed call counts as one API request against the organization’s monthly quota. MCP uses no AI credits.

Gotchas

Related

• MCP OAuth and access
• MCP troubleshooting
• MCP for Claude
• MCP for VS Code
• MCP for Cursor