A New Way To Connect Agents To VTX

VTX now exposes a dedicated CLI and MCP surface built around the same account, profile, bot, billing, runtime, and trading permissions already used inside the app.

That means an agent does not get a special backdoor into your account. It connects with an Agent Token, uses the scopes you grant, and still has to pass the same profile ownership checks, trading checks, billing checks, and runtime safety rules as normal VTX workflows.

In practice, this lets compatible agent apps discover VTX tools, check authentication, read profile and bot state, inspect AI configuration, view billing information, and, when explicitly scoped, perform controlled actions such as bot configuration or trade requests.

How It Works

The VTX CLI is the local command-line entrypoint. You authenticate once, then run commands against your VTX account.

For example, the CLI can:

• Check who you are authenticated as.
• Read profile and bot status.
• Configure or control bots when the token has the right scopes.
• Start a headless Client Mode runtime.
• Send market, limit, or cancel-order requests through the normal VTX trade paths.

MCP is the agent-app version of that same idea. Instead of an agent shelling out to individual commands, vtx-mcp starts a local MCP server. The agent app can then discover VTX tools and call them directly.

Under the hood, those tools still map back to the same shared VTX client methods and route checks as the CLI. A read-only token can read. A token without trading scope cannot trade. A token without bot-control scope cannot start or configure bots.

Safer Token And Runtime Handling

This release also tightens setup and safety behavior.

If the local token file is empty or malformed, the CLI now returns a clear setup message telling the user to run vtx auth login or remove the bad file and retry. That avoids confusing low-level JSON parsing errors during setup.

Runtime lease tokens are also redacted from CLI output. They remain in the local runtime state file where the runtime needs them, but they are not printed back into normal command output.

Better Compatibility With Agent Clients

The MCP server now supports both standard Content-Length stdio framing and newline-delimited JSON-RPC clients. It also replies using the same framing style the caller used.

That matters because different agent clients speak MCP over stdio slightly differently. This update makes VTX more tolerant of those real-world client shapes without changing the permission model.

The MCP tool schemas were also adjusted for better compatibility with agent model/tool parsers, so agents can inspect the available VTX tools without unnecessary schema friction.

Tested End To End

This release was validated with installed CLI/MCP builds outside the repository, direct MCP protocol checks, read-only and scoped mutation flows in development, production read-only checks, and an OpenClaw agent workflow.

The live development validation included tool discovery, authentication checks, missing-scope handling, scoped profile and bot operations, Client Mode runtime checks, and a small buy-then-sell trade test that returned the account to a flat position.

What This Means For Users

This is the foundation for using VTX Macro from agent apps and local automation without weakening account safety.

You can grant narrow access, keep tokens local, let agents inspect or operate within those limits, and rely on VTX’s existing backend checks to reject actions the token is not allowed to perform.