Innochip Nexus Developer Documentation
Build, execute, and scale autonomous AI workflows. Seamlessly connect Anthropic Claude Desktop, Cursor IDE, Windsurf, and Cloud LLMs directly to local sandboxed PC compute with zero host contamination.
🚀 60-Second Quickstart
Follow three simple steps to unlock sandboxed local code execution inside your favorite AI editor:
Download the Windows Node Installer
Grab the single-file setup package. It contains the silent Win32 service daemon, the ephemeral sandbox runtime, and the native MCP server engine.
irm https://innochipus.com/install.ps1 | iex
Runs instantly in user-space or admin, halts stale processes, auto-detects Google Drive, injects Claude/Cursor MCP config, and launches browser pairing.
Run Setup & Auto-Configure AI Clients
Execute the installer. It installs the background daemon and automatically detects and configures Claude Desktop and Cursor without overwriting your existing MCP servers.
Start-Process .\InnochipNodeSetup.exe -ArgumentList "/silent" -Wait
1-Click Hardware Binding
Upon completion, the installer automatically opens the secure pairing page with your machine's cryptographic GUID pre-filled. Simply click "Authorize & Bind This Device" to unlock tool execution.
🔌 Model Context Protocol (MCP) Architecture
Innochip Nexus Node implements the official Model Context Protocol (MCP) specification standard (JSON-RPC 2.0). Cloud and desktop AI agents connect directly to invoke real computing tools on your PC.
C:\ProgramData\Innochip\Node\sandbox\<uuid>) with 100% ephemeral teardown upon completion.
Supported Transports
stdio(Standard Input/Output): Recommended for local editors including Claude Desktop, Cursor, and Windsurf. Subprocess-based native IPC with sub-millisecond overhead.sse(Server-Sent Events / HTTP): Suitable for remote orchestration, team clusters, and cloud agent tunneling.
⚙️ Client Configuration Reference
While the v0.4.0 installer injects these configurations automatically, you can verify or inspect your editor's MCP settings manually at any time:
1. Claude Desktop (Anthropic)
Config file path: %APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"innochip-nexus-node": {
"command": "C:\\Program Files\\Innochip\\Node\\python\\python.exe",
"args": [
"-m",
"packages.nexus_mcp_v0_1.server",
"--stdio"
],
"cwd": "C:\\Program Files\\Innochip\\Node",
"env": {
"PYTHONUNBUFFERED": "1"
}
}
}
}
2. Cursor IDE
Config file path: %USERPROFILE%\.cursor\mcp.json
{
"mcpServers": {
"innochip-nexus-node": {
"command": "C:\\Program Files\\Innochip\\Node\\python\\python.exe",
"args": [
"-m",
"packages.nexus_mcp_v0_1.server",
"--stdio"
],
"cwd": "C:\\Program Files\\Innochip\\Node"
}
}
}
🛠️ MCP Tool Reference Catalog
The Nexus MCP Server exposes the following core tools to connected AI models:
| Tool Name | Access Level | Description |
|---|---|---|
| innochip_execute_sandboxed_python | Bound PC | Runs Python code inside a dedicated ephemeral sandbox with 30s watchdog watchdog and automated cleanup. |
| innochip_query_hardware_telemetry | Public | Returns real-time host telemetry: CPU load, RAM usage, GPU status, OS info, and active sandbox health. |
| innochip_get_device_status | Public | Queries local device pairing state, account email, subscription tier (Free/Pro/Studio), and hardware GUID. |
| innochip_read_sandboxed_file | Bound PC | Safely reads output or workspace files from an active sandbox session with strict anti-traversal checks. |
| innochip_write_sandboxed_file | Bound PC | Stages source or data files directly into an active sandbox workspace prior to execution. |
| innochip_cleanup_sandbox | Bound PC | Explicitly purges an ephemeral session directory, guaranteeing 0 bytes disk residue. |
🐍 Tool Deep Dive: execute_sandboxed_python
The primary execution engine. Enables Claude or Cursor to execute Python computations locally with strict isolation:
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "innochip_execute_sandboxed_python",
"arguments": {
"code": "import math\nprimes = [n for n in range(2, 50) if all(n % d != 0 for d in range(2, int(math.isqrt(n))+1))]\nprint(f'Computed {len(primes)} primes: {primes}')",
"timeout_seconds": 15.0
}
}
}
{
"jsonrpc": "2.0",
"id": 1,
"result": {
"content": [
{
"type": "text",
"text": "{\n \"status\": \"SUCCESS\",\n \"exit_code\": 0,\n \"stdout\": \"Computed 15 primes: [2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47]\\n\",\n \"stderr\": \"\",\n \"duration_ms\": 42.1,\n \"residue_bytes\": 0\n}"
}
],
"isError": false
}
}
🛡️ Sandbox Isolation Guarantees
Enterprise safety is the cornerstone of Innochip's design. We ensure:
- Ephemeral Workspaces: Every job executes in a uniquely generated UUID folder under
C:\ProgramData\Innochip\Node\sandbox\. It is completely obliterated upon exit. - Anti-Path-Traversal: The runtime strictly resolves all paths and aborts immediately if any input attempts
../traversal outside the sandbox boundary. - Watchdog Timeout: Tasks are capped at a maximum execution time (default 30 seconds). Runaway processes or unauthorized cryptomining loops are terminated forcefully.
- Hardware Zero-Trust: Devices are bound via immutable hardware hashes. Tasks cannot be hijacked by unauthenticated third parties.
❓ Developer FAQ & Troubleshooting
Why doesn't Claude Desktop see the Innochip tools?
Ensure you completely restart Claude Desktop after running the installer (Quit from the system tray, then relaunch). Claude only reads claude_desktop_config.json at initial startup.
Why do tools return "DEVICE_REGISTRATION_REQUIRED"?
Your local node is installed but not yet bound to an account. Open innochipus.com/pair and click "Authorize & Bind This Device". Binding takes 2 seconds and is 100% free for your primary PC.
How can I run the MCP server manually in terminal for debugging?
Run the following command in PowerShell to inspect stdio communication:
cd "C:\Program Files\Innochip\Node"
.\python\python.exe -m packages.nexus_mcp_v0_1.server --stdio
☁️ Cloud AI Bridge (Codex, ChatGPT & Google Drive)
While desktop editors (Claude Desktop, Cursor) connect directly over local IPC (stdio), Cloud-hosted AI agents (such as Codex, ChatGPT Plus/Team, or remote web agents) operate outside your local machine.
Innochip provides an automatic, zero-latency Google Drive Cloud Bridge that creates an asynchronous task mailbox on your PC.
How It Works
[Cloud AI / Codex]
│
▼ (writes task.json / script.py)
[Google Drive: My Drive\InnochipNode_<PC>\inbox\]
│
▼ (Innochip background node daemon polls every 1s)
[Local Ephemeral Sandbox (100% Isolated Execution)]
│
▼ (writes results, logs, and artifacts)
[Google Drive: My Drive\InnochipNode_<PC>\outbox\]
│
▼ (Cloud AI reads execution result)
[Cloud AI / Codex Completes Workflow]
Setup Steps
Install Google Drive for Desktop
Download and install the official Google Drive client. This maps a virtual cloud drive (typically G:\) onto your Windows workstation.
Automatic Gateway Detection
The Innochip Node installer and daemon automatically detect your Google Drive virtual drive and create the gateway directory: InnochipNode_<PC_NAME> containing inbox/ and outbox/.
Multi-Account Compatibility Note
It does not matter if your PC's Google Drive is signed in with a different Google account than your Innochip account. As long as Google Drive for Desktop is installed locally, the sync folder functions as your secure computing gateway.
🪐 Google Antigravity MCP Integration
Google Antigravity autonomous coding agents natively communicate with Innochip Nexus Node through standard Model Context Protocol tool declarations.
Antigravity Configuration
Add the following block to your global mcp_config.json (located at %USERPROFILE%\.gemini\config\mcp_config.json):
{
"mcpServers": {
"innochip_nexus_node": {
"command": "python",
"args": [
"C:\\Program Files\\Innochip\\Node\\innochip_mcp_server.py",
"--stdio"
],
"env": {
"PYTHONUNBUFFERED": "1"
},
"disabled": false
}
}
}
The 1-line installer automatically writes tool schemas into %USERPROFILE%\.gemini\antigravity\mcp\innochip_nexus_node\ so that Antigravity agents can immediately discover and call innochip_execute_sandboxed_python and innochip_query_hardware_telemetry.
🖥️ Desktop GUI Control Panel
For users who prefer a graphical dashboard over terminal commands, every Innochip Node installation includes the Innochip Desktop Control Panel.
Key Capabilities of the Desktop Control Panel
- Live Hardware & Binding Monitor: View machine GUID, account tier, and zero-trust verification status.
- AI Bridge Status Checker: Visual indicators showing whether Claude, Cursor, Antigravity, and Google Drive are connected.
- One-Click Folder Navigation: Instantly open your
inboxandoutboxfolders in Windows File Explorer. - Self-Diagnostic Engine: Run on-demand self-tests to verify Python runtime, sandbox permissions, and network endpoints.
- Real-Time Logs: Stream the last 100 lines of daemon telemetry and task execution history.
Launch via Windows Start Menu → Innochip Control Panel, or run:
python "C:\Program Files\Innochip\Node\InnochipControlPanel.py"
🔒 Windows SmartScreen & Code Signing FAQ
Why did Windows SmartScreen show "Unknown Publisher"?
Microsoft Defender SmartScreen flags any newly compiled executable that lacks an Extended Validation (EV) Code Signing Certificate, regardless of how safe the program is.
EV Code Signing certificates cost approximately $400 to $800 per year, require formal corporate incorporation audits, and mandate physical cryptographic USB tokens. For agile developers and open-source infrastructure, this creates an unnecessary artificial barrier.
Why the PowerShell 1-Liner is the Recommended Standard
irm https://innochipus.com/install.ps1 | iex
Unlike a closed-source binary, the PowerShell script is 100% transparent and auditable in plain text. Anyone can visit innochipus.com/install.ps1 in a browser to inspect every single operation before running it. It never triggers SmartScreen warnings and sets up everything in under 10 seconds.
* For enterprise customers who require signed enterprise installers, officially signed EV binaries will be available in Phase 3.