The Conway Terminal

The Conway Terminal gives AI agents write access to the real world. With a single install, any AI agent -- Claude, Cursor, ChatGPT, or your own -- can create cloud sandboxes, execute code, deploy applications, expose services, and pay for compute with USDC. No human intermediary required.

It implements the Model Context Protocol (MCP) standard, so it works out of the box with any MCP-compatible client.

What It Gives Your Agent

Cloud Sandboxes -- Isolated Linux VMs, provisioned in seconds
Code Execution -- Run any command or script
File Operations -- Read and write files in sandboxes
Interactive Sessions -- PTY sessions for REPLs, editors, and interactive CLIs
Port Forwarding -- Expose services to the internet with automatic SSL
x402 Payments -- Pay for compute with USDC, no login or KYC required

Features

Sandbox Management

Create and manage isolated Linux VMs with configurable resources:

1-4 vCPUs
512MB - 8GB RAM
1-50GB disk
Multiple regions (EU, US)

PTY Sessions

Interactive pseudo-terminal sessions for:

Python/Node.js REPLs
Text editors (vim, nano)
Interactive CLIs
Long-running shell sessions

x402 Payments

Built-in support for x402 permissionless payments:

Automatic wallet management
USDC payments on Base network
No login, no KYC, no human approval required

Quick Start

Installation

curl -fsSL https://conway.tech/terminal.sh | sh

Or install via npm:

npm install -g conway-terminal

Configuration

Set your API key:

export CONWAY_API_KEY="your-api-key"
export CONWAY_API_URL="https://api.conway.tech"  # Optional, defaults to this

Running the Server

Stdio mode (for Claude Desktop, Cursor):

conway-terminal

HTTP mode (for web integrations):

conway-terminal --http 3000

Integration Examples

Claude Desktop

Add to your Claude Desktop configuration (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "conway": {
      "command": "conway-terminal",
      "env": {
        "CONWAY_API_KEY": "your-api-key"
      }
    }
  }
}

Cursor

Add to your Cursor MCP settings:

{
  "mcpServers": {
    "conway": {
      "command": "npx",
      "args": ["conway-terminal"],
      "env": {
        "CONWAY_API_KEY": "your-api-key"
      }
    }
  }
}

Programmatic Usage

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";

const transport = new StdioClientTransport({
  command: "conway-terminal",
  env: { CONWAY_API_KEY: "your-api-key" },
});

const client = new Client({ name: "my-app", version: "1.0.0" });
await client.connect(transport);

// List available tools
const tools = await client.listTools();

// Create a sandbox
const result = await client.callTool("sandbox_create", {
  name: "my-sandbox",
  vcpu: 2,
  memory_mb: 2048,
});

How It Works

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│   AI Agent      │────▶│ Conway Terminal  │────▶│  Conway API     │
│ (Claude/Cursor) │     │                 │     │                 │
└─────────────────┘     └─────────────────┘     └─────────────────┘
                                                        │
                                                        ▼
                                               ┌─────────────────┐
                                               │  Cloud Sandbox  │
                                               │  (Isolated VM)  │
                                               └─────────────────┘

Your AI agent connects to the Conway Terminal, which translates tool calls into Conway API requests. The API provisions isolated Linux VMs where your agent can execute code, deploy applications, and expose services to the internet.

Source Code

GitHub: github.com/conwayresearch/conway-terminal
npm: conway-terminal

On this page