MCP server

@tronchartsxyz/mcp-server exposes the provider API as MCP tools and prompts, so an agent operates your firm by calling tools rather than composing HTTP requests. Your credential lives in the server’s environment; agent keys never leave the process.

You can also build the server from the monorepo and point your MCP client at the built entry point:

cd mcp-server && bun install && bun run build

"command": "node",
"args": ["/absolute/path/to/mcp-server/dist/index.js"]

Everything else on this page — the env contract, the tools, the transports — is exactly the same either way.

Configure

Claude Desktop, Claude Code, or Cursor:

{
  "mcpServers": {
    "troncharts": {
      "command": "npx",
      "args": ["-y", "@tronchartsxyz/mcp-server"],
      "env": {
        "TRON_CHARTS_API_KEY": "your-api-key",
        "TRON_CHARTS_API_SECRET": "your-api-secret",
        "TRON_CHARTS_BASE_URL": "https://api.troncharts.xyz"
      }
    }
  }
}

Default transport is stdio. Set TRON_CHARTS_MCP_TRANSPORT=http (with TRON_CHARTS_MCP_PORT and TRON_CHARTS_MCP_TOKEN) to host it over HTTP instead.

Firm-lifecycle tools

Tool	Does
`list_firms` / `get_firm`	Read your firms.
`create_firm`	Create a firm (`name`, `multisigAddress`, `feeConfig`, …).
`check_firm_readiness`	The go-live checklist.
`deploy_prop_instance`	Dry-run or broadcast the deploy.
`set_firm_status`	Go live, pause, migrate.
`list_payouts` / `approve_payout`	Work the payout queue.

The trading and liquidation tool sets are filtered by your credential’s tier. The firm-lifecycle tools above are listed at every tier and gated server-side instead: a credential without firm:operate gets a 403 on the call. Either way an agent cannot reach past what your credential is allowed to do, which is the point.

The one-shot prompt

The server ships a scripted launch-firm prompt that runs the whole lifecycle — create → readiness → dry-run deploy → broadcast → re-check → go live — from two inputs (name, multisigAddress). Invoke it and approve each step.

Streamable HTTP transport

The platform also exposes an MCP endpoint over Streamable HTTP / SSE for clients that speak the protocol directly rather than spawning a local server. The tool catalog is tier-filtered per session, exactly as with stdio.

That endpoint carries the read-only, liquidation and trading tools only — the firm-lifecycle tools on this page are registered by the stdio server, so run that one to operate a firm. The launch-firm prompt is offered on both transports, but over Streamable HTTP every step it scripts calls a tool that is not there.

MCP or plain REST?

MCP when an agent should operate the firm live — watching readiness, deploying, working the payout queue.
REST or the SDK when you want code you own and run yourself.

Both share the same credential and scopes.