MCP server

TronGasAi ships an MCP server — a thin wrapper over the API that exposes actions as tools an AI assistant (e.g. Claude Desktop) can call. It authenticates with an API key, so everything is bounded by that key’s scope and daily spend limit.

Tools

• Read — get_pricing, get_balance, preview_*, list_* (work with any key).
• Write (spends TRX) — buy_energy, create_auto_refill, enable_smart, pause_automation / resume_automation / cancel_automation (need a purchase or full key).

Each write tool is labelled with the scope it needs. A read-only key lets an agent monitor an account with no ability to spend; add a daily spend limit to cap a purchase key.

Recommended setup for agents

1. Create a dedicated API key from the dashboard with the least scope the agent needs (read for monitoring, purchase to act).
2. Set a daily spend limit appropriate to the task.
3. Point your MCP client at the server (remote or stdio, see below) using that key. Rotate or revoke it anytime from the dashboard.

Transport

Remote (recommended) — https://mcp.trongas.ai/mcp

No install needed. Add it as a remote MCP server in your client and authenticate with a request header — either Authorization: Bearer <your-api-key> or X-Api-Key: <your-api-key>. The server is stateless: it doesn’t store your key, it reads it fresh from the header on every call.

Note this is header-based auth, not an OAuth “connect” flow — your MCP client needs to support sending a custom header, not just a one-click account link.

stdio (local)

Install the package and run it as a local process (e.g. Claude Desktop config):

npm install && npm run build
node dist/index.js

Configure via env vars: TRONGASAI_API_KEY (your key) and optionally TRONGASAI_API_URL (defaults to https://trongas.ai/api/v1).