Local Commander

Local Commander for Cursor, DeepSeek Harness, Antigravity, Kiro and LM Studio

Connects Cursor, DeepSeek Harness, Antigravity, Kiro and LM Studio to Local Commander at https://localcmd.app/mcp. Your AI gets hands on every computer you own: it can run commands and scripts, change settings, edit files and check local services on any of them, from one chat. Your workspace at localcmd.app shows which computer ran each request and how it ended.

This repository holds configuration only: a Cursor plugin, a DeepSeek Harness bundle, an Antigravity plugin, a Kiro Power and LM Studio settings. The service runs at localcmd.app.

Requirements

• A Local Commander account at localcmd.app.
• A Windows 11 x64 computer with Node.js 20 or later and the paired agent: run npx local-commander-agent@latest on it, then choose full access or selected folders.
• The computer must be online for work and for reading results.

Once you approve a computer and connect your AI, it can do what your Windows account can do, just like you at the keyboard. You can limit it to certain folders, block commands, stop running work, or switch a computer off from the site at any time. Every run shows which computer did it. Whatever the AI asks for goes back to the AI.

Cursor

1. Install the plugin, or add this to ~/.cursor/mcp.json:

   {
     "mcpServers": {
       "local-commander": { "url": "https://localcmd.app/mcp" }
     }
   }
   

2. Cursor opens localcmd.app to sign in and approve access (OAuth 2.0). Sign-in happens on localcmd.app; Cursor never sees your password.

3. Ask the agent to list your computers, then work on an approved project.

Troubleshooting:

• The agent reports no computers: start the agent on the Windows computer and check that it shows as online at localcmd.app.
• Sign-in loops or fails: open Output, choose "MCP Logs", and check the OAuth error; then remove and re-add the server.
• A result is unknown: the computer dropped during the task. The work is not re-run automatically; check the computer and ask again.

DeepSeek Harness

A configuration-only DeepSeek Harness bundle. The tools appear as mcp__local-commander__<tool>.

The harness MCP client sends static headers only, so the bundle starts mcp-remote over stdio. On first use it opens localcmd.app in your browser to sign in and approve access (OAuth 2.0), then keeps the token refreshed. In your localcmd.app workspace these requests show the client name "MCP CLI Proxy", the name mcp-remote registers.

DeepSeek Harness asks for Node.js 22.19 or later on the machine where you chat.

Install: download local-commander-dsh-1.0.0.tgz from the latest release, then run:

dsh plugin --profile web add ./local-commander-dsh-1.0.0.tgz
dsh web

You can also build the tarball from this repository with npm pack.

Remove:

dsh plugin --profile web remove local-commander-dsh

Antigravity

This repository is also an Antigravity plugin (.antigravity-plugin/plugin.json plus mcp_config.json). Like the DeepSeek Harness bundle, it starts mcp-remote over stdio, which opens localcmd.app in your browser to sign in and approve access (OAuth 2.0) and keeps the token refreshed. Requests show the client name "MCP CLI Proxy" in your localcmd.app workspace.

Install:

agy plugin install https://github.com/netbka/local-commander-plugins

Or clone the repository and run agy plugin install <folder>. Remove with agy plugin uninstall local-commander.

Kiro

The kiro-power/ folder is a Kiro Power in the Agent Plugins format (plugin.json and mcp.json). It points Kiro at the remote server https://localcmd.app/mcp (Streamable HTTP), which uses OAuth 2.0 sign-in at localcmd.app.

Install: in Kiro, open the Powers panel, choose Add Custom Power, then Import power from GitHub, and enter https://github.com/netbka/local-commander-plugins/tree/main/kiro-power.

LM Studio

LM Studio 0.3.17 or later runs MCP servers listed in its mcp.json. Like the DeepSeek Harness bundle, the entry starts mcp-remote over stdio, which opens localcmd.app in your browser to sign in and approve access (OAuth 2.0) and keeps the token refreshed. Requests show the client name "MCP CLI Proxy" in your localcmd.app workspace. The computer running LM Studio needs Node.js 20 or later.

1. In LM Studio, open the Program tab in the right sidebar, choose Install > Edit mcp.json, and add:

   {
     "mcpServers": {
       "local-commander": {
         "command": "npx",
         "args": ["-y", "[email protected]", "https://localcmd.app/mcp"]
       }
     }
   }
   

   Or paste this link into your browser's address bar on that computer, and LM Studio adds the same entry:

   lmstudio://add_mcp?name=local-commander&config=eyJjb21tYW5kIjoibnB4IiwiYXJncyI6WyIteSIsIm1jcC1yZW1vdGVAMC4xNS4xIiwiaHR0cHM6Ly9sb2NhbGNtZC5hcHAvbWNwIl19
   

2. In a chat, turn on mcp/local-commander and choose a model that supports tool calls. On first use your browser opens localcmd.app to sign in and approve access.

3. Ask the model to list your computers, then work on an approved project.

Links

Privacy: https://localcmd.app/privacy. Terms: https://localcmd.app/terms. Support: [email protected] (https://localcmd.app/support).

License

MIT, for the files in this repository.