UniRate API MCP server

UniRate MCP Server

A Model Context Protocol server for the UniRate API — give Claude, Cursor, Continue, and any MCP-compatible AI assistant first-class access to currency conversion and exchange rates.

• 🔄 Real-time conversion between 170+ currencies (fiat + major crypto)
• 📈 Historical rates back to 1999 (Pro plan)
• 🆓 Free tier, no credit card required — get a key at unirateapi.com
• 🧩 Four tools, fully-typed inputs (Zod schemas), structured outputs
• 🌐 Stdio + Streamable HTTP/SSE transports — run locally or host as a remote MCP endpoint
• ⚡ Pure Node 18+, single dependency on @modelcontextprotocol/sdk

Why this exists

Most "currency for AI" workflows today involve hand-rolled fetch wrappers in custom tools, or generic HTTP MCP servers that hand the model raw JSON. This server gives models a tight, typed, currency-aware tool surface — they ask "what was 100 USD in EUR on 2020-03-15?" and get back a formatted answer plus a structured payload they can chain into other tool calls.

Quick start

1. Install

npm install -g @unirate/mcp

Or run on demand with npx @unirate/mcp (no install).

2. Get a UniRate API key

Free tier covers convert, latestrate, and listcurrencies. Sign up at unirateapi.com — no credit card required.

3. Wire it into your MCP client

Claude Desktop

Edit ~/Library/Application Support/Claude/claudedesktopconfig.json (macOS) or %APPDATA%\Claude\claudedesktopconfig.json (Windows):

{
  "mcpServers": {
    "unirate": {
      "command": "npx",
      "args": ["-y", "@unirate/mcp"],
      "env": {
        "UNIRATE_API_KEY": "your-api-key-here"
      }
    }
  }
}

Restart Claude Desktop. The four UniRate tools will appear in the tool picker.

Cursor / Continue / Cline

Add to your MCP config (.cursor/mcp.json, ~/.continue/config.json, etc.):

{
  "mcpServers": {
    "unirate": {
      "command": "npx",
      "args": ["-y", "@unirate/mcp"],
      "env": { "UNIRATE_API_KEY": "your-api-key-here" }
    }
  }
}

From source

git clone https://github.com/UniRate-API/unirate-mcp.git
cd unirate-mcp
npm install && npm run build
UNIRATE_API_KEY=your-key node dist/index.js

4. Run as a remote endpoint (Streamable HTTP / SSE)

By default the server uses stdio, which is what Claude Desktop and most MCP clients want. To host it as a remote endpoint instead — for shared use, multi-user deployments, or browser-based clients — start it in HTTP mode:

UNIRATE_API_KEY=your-key unirate-mcp --http 3001
# or via env:
UNIRATE_API_KEY=your-key UNIRATE_MCP_HTTP_PORT=3001 unirate-mcp

That exposes:

• POST /mcp — Streamable HTTP endpoint (SSE-capable). Stateless: a fresh server is built per request, so the same process can serve many concurrent clients.
• GET /healthz — JSON liveness probe ({ "status": "ok", "server": "unirate-mcp", "version": "..." }).

Point any Streamable-HTTP-capable MCP client (Claude Desktop with remote server support, Cursor remote MCP, etc.) at http://your-host:3001/mcp. Drop it behind a reverse proxy + TLS for production.

Docker

Multi-arch images (linux/amd64, linux/arm64) are published to both the GitHub Container Registry (ghcr.io/unirate-api/unirate-mcp) and Docker Hub (unirate/unirate-mcp) — use whichever you prefer:

# stdio mode (for local AI clients — pipe stdin/stdout)
docker run --rm -i -e UNIRATE_API_KEY="your-key" \
  ghcr.io/unirate-api/unirate-mcp:latest
# ...or from Docker Hub:

Shortened. The full README is on GitHub.