MCP server

What MCP is

The Model Context Protocol (MCP) is an open standard that lets an AI app call tools on a server: the app sends a structured request, the server answers with data, and the model writes its reply from that data instead of from memory. FundFacts runs a remote MCP server at https://fundfactsapi.com/api/mcp (Streamable HTTP). Connect it once and Claude, ChatGPT, Claude Code, Cursor, VS Code, Windsurf or Codex can look up any fund or ETF by ISIN or name, compare funds, look through a portfolio, measure holdings overlap and read French SCPIs, with the figures the fund houses publish and the date of those figures.

It works on every FundFacts plan, Free included, and takes about a minute to connect. Create a free account (no card), then follow the steps for your app below.

What you can ask

Ask in plain language; the assistant picks the tool, calls it and quotes the figures with their as-of date.

The seven tools

All seven tools only read data. Each fund found counts as one request, exactly like the REST API; misses, searches and the example fund are not counted.

The tour prompt and the example fund

The server also offers an MCP prompt, tour (“Tour of FundFacts”). Pick it from the app's prompt menu, or just ask “what can FundFacts do?”: the assistant explains the service in a few bullets, then calls get_fund on the example fund IE00B4L5Y983 (iShares Core MSCI World UCITS ETF) and walks through what it costs, how risky it is, what it holds and how it performed.

The example fund is preloaded, answers at once and is not counted as a request, so a new account sees a real answer before spending anything. The prompt takes an optional isin (or a fund name) to run the tour on your own fund instead.

Cards in ChatGPT and Claude

In ChatGPT and in Claude on the web and desktop, results render as interactive cards (MCP Apps) next to the answer: a fund card, a comparison table, a portfolio look-through, a holdings overlap and an SCPI card. The model still receives the data, so it can answer follow-up questions from the same call. Claude Code shows the result as text.

Illustration: a fund card for IE00B4L5Y983 in Claude.

Illustration: a portfolio look-through card in Claude.

Setup: Claude on the web, desktop and Cowork

The connector follows your Claude account, so it is also available in Claude Desktop and Cowork.

Setup: ChatGPT

Remote MCP servers are added in developer mode, on Plus, Pro, Business, Enterprise and Edu, on the web.

ChatGPT cannot send an API key, so it always signs in with the account.

Setup: Claude Code

claude mcp add --transport http fundfacts https://fundfactsapi.com/api/mcp

Then run /mcp in a Claude Code session, pick fundfacts and sign in with your FundFacts account. To use an API key instead of the sign-in:

claude mcp add --transport http fundfacts https://fundfactsapi.com/api/mcp --header "Authorization: Bearer ffk_your_api_key"

Or install the FundFacts plugin, which adds the server and four skills that tell Claude Code how to read and present the fund data (public repository: github.com/fundfactsapi/fundfacts-plugin), then /mcp to sign in:

/plugin marketplace add fundfactsapi/fundfacts-plugin
/plugin install fundfacts@fundfacts

Setup: Cursor

Create or extend .cursor/mcp.json (or ~/.cursor/mcp.json for every project), then enable the tools in Cursor's MCP settings:

.cursor/mcp.json — json
{
  "mcpServers": {
    "fundfacts": {
      "url": "https://fundfactsapi.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ffk_your_api_key"
      }
    }
  }
}

Setup: VS Code (GitHub Copilot agent mode)

Save as .vscode/mcp.json; VS Code asks for the key once and keeps it out of the file:

.vscode/mcp.json — json
{
  "inputs": [
    {
      "id": "fundfacts-key",
      "type": "promptString",
      "description": "FundFacts API key",
      "password": true
    }
  ],
  "servers": {
    "fundfacts": {
      "type": "http",
      "url": "https://fundfactsapi.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ${input:fundfacts-key}"
      }
    }
  }
}

Setup: Windsurf

Add the server to ~/.codeium/windsurf/mcp_config.json; Windsurf names the URL key serverUrl:

mcp_config.json — json
{
  "mcpServers": {
    "fundfacts": {
      "serverUrl": "https://fundfactsapi.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ffk_your_api_key"
      }
    }
  }
}

Setup: Codex

Codex reads MCP servers from ~/.codex/config.toml; export FUNDFACTS_API_KEY=ffk_… in the shell that starts it:

~/.codex/config.toml — toml
[mcp_servers.fundfacts]
url = "https://fundfactsapi.com/api/mcp"
bearer_token_env_var = "FUNDFACTS_API_KEY"

Setup: any other MCP client

Clients that support remote servers take the URL and a header, usually in this shape:

json
{
  "mcpServers": {
    "fundfacts": {
      "url": "https://fundfactsapi.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ffk_your_api_key"
      }
    }
  }
}

Clients that only speak stdio (including claude_desktop_config.json) go through the mcp-remote bridge:

json
{
  "mcpServers": {
    "fundfacts": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://fundfactsapi.com/api/mcp",
        "--header",
        "Authorization: Bearer ffk_your_api_key"
      ]
    }
  }
}

Authentication

Sign in with your FundFacts account (OAuth). ChatGPT, claude.ai, Claude Desktop, Cowork, and Claude Code through /mcp. A consent screen opens in the browser; the app gets an access token (OAuth 2.1 with PKCE) and no API key leaves your dashboard. The token only works on the MCP server, never on the REST API.

An API key as a header. Claude Code, Cursor, VS Code, Windsurf, Codex and scripts send Authorization: Bearer ffk_…, the same key as the REST API (Authentication). Create a dedicated key per tool so you can revoke one without touching the others.

Limits and billing

Privacy and disconnecting

Troubleshooting

FAQ

Also see AI agents & MCP for llms.txt, the OpenAPI spec and the AI companion kit for coding agents.