nexalware-mcp

Nexalware as an MCP server, over stdio, for Python. Point any MCP-aware host (Claude Desktop, Claude Code, Cursor, etc.) at it, and your device control tools show up automatically, the host's own model decides when to call them. It's a thin wrapper over the Python SDK's make_tools(), same operations, same permissions, just discoverable instead of hand-coded.

Already on Node? See @nexalware/mcp, same tool set. Building your own agent instead of using a pre-built host? Use the SDK directly, in Python or TypeScript. It's one function call away in code you already control, no subprocess or protocol discovery needed.

Get an API key

Create one on the dashboard (API Keys), scoped with a DeviceGrant to only the device(s) this agent should touch. This key ends up sitting in a local config file, least privilege matters more here than for a typical server-side integration.

Claude Code

claude mcp add nexalware -e NEXALWARE_API_KEY=nxw_live_sk_your_key_here -- uvx nexalware-mcp

Claude Desktop, or any host using a mcpServers config file

Add this block to the host's MCP config (Claude Desktop: Settings -> Developer -> Edit Config):

{
  "mcpServers": {
    "nexalware": {
      "command": "uvx",
      "args": ["nexalware-mcp"],
      "env": {
        "NEXALWARE_API_KEY": "nxw_live_sk_your_key_here"
      }
    }
  }
}

Restart the host afterward, tools only load at startup. This uses uvx to run the package without a separate install step, every launch runs the latest published release. No uv? Install it once (curl -LsSf https://astral.sh/uv/install.sh | sh), or fall back to a regular install:

pip install nexalware-mcp

{
  "mcpServers": {
    "nexalware": {
      "command": "nexalware-mcp",
      "env": { "NEXALWARE_API_KEY": "nxw_live_sk_your_key_here" }
    }
  }
}

Environment variables

Tools

Every tool takes and returns the same shape as its equivalent Python SDK method, a failed call comes back as a normal MCP tool error with the same message the API itself returns, not a crash.

What it can't do

Nothing outside the calling key's own DeviceGrant. There's no tool to create a grant, register a device, or manage API keys, those stay dashboard-only, an MCP tool call is only ever as capable as the key you gave it, never able to expand its own access.

create_schedule/update_schedule fire existing commands from a device's catalog, they can't invent new ones. Defining, editing, or deleting a command definition itself stays a dashboard-only action.

Links

• Full docs
• SDK Reference (Python) / SDK Reference (TypeScript) - the same operations, called directly from your own code.
• Device Orchestration - what list_sub_devices/send_sub_device_command actually operate on.

License

MIT