Skip to content

For AI agents#

Anboto ships first-party tooling that connects AI agents (Claude, Cursor, Codex, or any MCP-compatible client) to the Trading API — including the execution algos (TWAP, VWAP, POV, IS, ICEBERG) that make Anboto an execution desk rather than an order button.

Everything lives in the open-source anboto-agent-kit (MIT):

  • @anboto/mcp — MCP server (stdio) with 22 tools: order creation and monitoring, portfolio, market data, preflight validation and guardrails
  • @anboto/core — typed TypeScript client (HMAC/RSA signing included)
  • anboto-trading — a Claude Skill teaching agents algo selection, execution monitoring and TCA

Quickstart (Claude Code)#

claude mcp add anboto \
  -e ANBOTO_API_KEY=your-key \
  -e ANBOTO_API_SECRET=your-base64-secret \
  -- npx -y @anboto/mcp --testnet

Claude Desktop / Cursor (JSON config):

{
  "mcpServers": {
    "anboto": {
      "command": "npx",
      "args": ["-y", "@anboto/mcp", "--testnet"],
      "env": {
        "ANBOTO_API_KEY": "your-key",
        "ANBOTO_API_SECRET": "your-base64-secret"
      }
    }
  }
}

Then ask the agent: "On testnet, buy 0.001 BTC on Binance via TWAP over 5 minutes, monitor to completion, and give me an execution report."

Guardrails#

API keys stay on your machine (env vars only — never in the model context). Optional flags, all fail-closed:

Flag Effect
--testnet Simulated environment (recommended to start)
--read-only Trading tools are not registered at all
--max-order-notional=50000 Reject orders above an estimated notional
--allowed-exchanges=BINANCE,OKX Exchange allowlist
--allowed-symbols=BTC/USDT Symbol allowlist

Machine-readable docs#