Connect your AI to your machines

infersurge speaks MCP (Model Context Protocol) — the same open standard used by ChatGPT, Claude, VS Code, Cursor and JetBrains. One endpoint, one token, every device you own.

STEP 01

Add a device

Sign in at app.infersurge.com, click Add Device, and run the one-line command on the machine you want to control. It connects out to the gateway over encrypted WebSocket — no inbound ports, no VPN.

STEP 02

Copy your MCP token

In the console's MCP card, copy the endpoint URL and generate your personal MCP token. The token authenticates your AI client to the gateway; revoke or rotate it anytime.

STEP 03

Point your AI at it

Add the remote MCP server to your client of choice below. Your AI instantly sees every online device as a set of tools — terminal, files, search, transfers, desktop control.

Per-client setup guides

All guides use the same two values from your console: the endpoint https://api.infersurge.com/mcp and your MCP token. Pick your harness.

ChatGPT Connectors · remote MCP

Works with ChatGPT Plus / Pro / Business via custom connectors (developer mode).

  1. Open Settings → Connectors and enable Developer mode (available on paid plans).
  2. Click Create / Add custom connector.
  3. Set the MCP Server URL to https://api.infersurge.com/mcp.
  4. Authentication: API key / Bearer token, paste your infersurge MCP token.
  5. Create, then start a new chat and pick your connector under the Developer mode tools menu.
Tip: tool names are prefixed per device (e.g. web1__list_directory) so one connector covers your whole fleet.

Claude Web & Desktop · custom connector

Works on claude.ai and Claude Desktop via remote MCP connectors.

  1. Open Settings → Connectors → Add custom connector (claude.ai, or Claude Desktop → Settings → Connectors).
  2. Remote MCP server URL: https://api.infersurge.com/mcp.
  3. Authentication: choose Custom headers and add Authorization: Bearer <your-mcp-token>.
  4. Save and enable the connector in your conversation (plug icon).
Note: Claude Desktop also supports local stdio servers — if you prefer, run a local MCP proxy that forwards to the same endpoint; the remote setup above is simpler.

VS Code Copilot agent mode · mcp.json

GitHub Copilot agent mode speaks MCP natively — no extension needed.

  1. Command palette → MCP: Add Server → HTTP (HTTP or Server-Sent Events), or edit .vscode/mcp.json directly with the config below.
  2. Reload and start Copilot agent mode; the infersurge tools appear in the tools picker.
// .vscode/mcp.json — user scope works too
{
  "servers": {
    "infersurge": {
      "type": "http",
      "url": "https://api.infersurge.com/mcp",
      "headers": { "Authorization": "Bearer <your-mcp-token>" }
    }
  }
}

Cursor MCP · mcp.json

Same remote-MCP shape as VS Code.

  1. Open Cursor Settings → MCP & Integrations → Add Custom MCP (or edit ~/.cursor/mcp.json).
  2. Use the config below, then toggle the server on.
{
  "mcpServers": {
    "infersurge": {
      "url": "https://api.infersurge.com/mcp",
      "headers": { "Authorization": "Bearer <your-mcp-token>" }
    }
  }
}

JetBrains IDEs AI Assistant · MCP

IntelliJ, PyCharm, GoLand, WebStorm — AI Assistant has built-in MCP client support.

  1. Settings → Tools → AI Assistant → MCP (or MCP Servers), click Add.
  2. Type: URL; URL: https://api.infersurge.com/mcp.
  3. Add request header Authorization: Bearer <your-mcp-token>.
  4. Enable the server and ask the assistant to use your devices.
JetBrains late versions also read .jetbrains/mcp.json in the project root with the same shape as the Cursor example above.

Any other MCP client Streamable HTTP

infersurge is a standard MCP Streamable-HTTP server — anything that can open a URL can connect.

ValueSetting
Endpointhttps://api.infersurge.com/mcp
TransportStreamable HTTP (JSON-RPC over POST)
AuthAuthorization: Bearer <mcp-token>
ToolsAggregated per device: <device>__<tool>
Billing: every successful tools/call counts as one command against your plan; failed, offline and timed-out calls are automatically refunded. Screenshots and image reads are metered by payload size.

The device agent: one binary

The isg CLI is a single static Go binary — no runtime, no Node, no dependencies. It enrolls your machine, holds a persistent encrypted WebSocket to the gateway, and executes tool calls locally under your path whitelist.

# Linux (amd64) — other platforms: linux-arm64, darwin-amd64/arm64, windows-amd64/arm64
curl -fsSL https://www.infersurge.com/isg/install.sh | bash

isg connect --server wss://api.infersurge.com/api/devices/connect --token <one-time-token>

Generate the exact command (token included, expires in 10 minutes) from the console's Add Device dialog. Upgrade anytime with isg upgrade — updates are SHA-256 verified and swap atomically.

Troubleshooting

The three things that cover 95% of first-connection issues.

SymptomFix
401 on connectToken missing or revoked — regenerate in the console MCP card and update the Authorization header.
Tools list emptyNo device online — re-run isg connect on the target machine and check it shows online in the console.
429 on tool callMonthly command quota reached — check usage in the console or upgrade the plan.
Device shows offlineCLI process stopped or network blocked outbound WSS (tcp/443) — restart it; it reconnects with exponential backoff.

Two minutes from signup to a controlled machine

Free plan: 1 device, 10,000 commands / month. No credit card.

Open the console Back to overview