Local MCP: without an account or through the desktop app
AI clients that start MCP servers as a local program can use WECOMMANDER without the cloud. There are two local modes: the agent alone, limited to folders you name, and the desktop app's local endpoint, which gives local AI clients your saved SSH servers.
1. The agent in local mode (no account)
webrising-commender mcp-stdio --allowed-root "$HOME/projects"| Option | Default | Meaning |
|---|---|---|
| --allowed-root <folder> | required, repeatable | The only folders the tools may touch; everything else is refused, including through symbolic links. |
| --allow-writes | off | File writes: write, append, create folders, move, replace, delete, undo. |
| --allow-commands | off | Commands, background tasks and terminals. |
| --audit-log <file> | none | One JSON line per call: time, tool, risk, outcome. File contents are never written. |
| --no-redact | off | Turns off masking of secrets in command, task, terminal and process output. |
Without --allow-writes and --allow-commands the AI can only read and search. With both it works like Desktop Commander inside the allowed folders and still keeps undo snapshots of every write. Example for Claude Desktop (Settings → Developer → Edit Config):
{
"mcpServers": {
"wecommander": {
"command": "/opt/wecommander/bin/webrising-commender",
"args": ["mcp-stdio", "--allowed-root", "/home/you/projects", "--allow-writes", "--audit-log", "/home/you/.wecommander-local.log"]
}
}
}On Windows use C:\Program Files\WECOMMANDER\webrising-commender.exe and Windows folders; on macOS the binary is /usr/local/bin/webrising-commender. The command waits silently for a client on stdin; that is expected.
2. The desktop app's local endpoint (Windows)
- Settings → AI in the Windows app can add WECOMMANDER to Claude Desktop, Claude Code, Cursor and Codex. It shows the exact change to the client's config file first, keeps every other entry, and saves a .bak copy.
- The client starts wecommander-core --mcp-stdio, a thin bridge to the running app. If the app is closed, the bridge answers that WECOMMANDER must be opened.
- Each new AI client is shown once and must be allowed; you can revoke it later.
- Only servers whose AI access has “AI on this computer” turned on are visible. Commands go straight from your computer to the server and never pass through wecommander.app.
- Commands need your approval in the app (single use, 5 minutes, bound to the exact command and server), unless the server is read-only (no commands) or on timed autonomy (up to 8 hours).
- A local audit file records time, client, tool, server, risk and outcome; never command text, output or file contents.
The macOS app does not have this settings page yet, and the Linux Management Center does not offer local MCP.
Local or remote?
| Local mode | Remote (wecommander.app/mcp) | |
|---|---|---|
| AI clients | Desktop clients that start a local program | Any MCP client, including claude.ai, ChatGPT and mobile |
| Machines | This computer (agent) or your saved SSH servers (app) | Every device and shared SSH server, one URL |
| High-risk approval | The flags you started it with, or the app's approval | Single-use approval per call, trust levels, allowed commands |
| Audit | Local file | Central audit log in the dashboard |
- AI access with MCP: how it works: What MCP (Model Context Protocol) is, the one WECOMMANDER MCP URL (https://wecommander.app/mcp), how a tool call reaches your machine, and the two ways to connect a server: the agent or SSH.
- Hybrid Access: AI on SSH servers without an agent: Share saved SSH servers with AI clients through WECOMMANDER on your own computer (Windows, macOS, Linux): nothing installed on the server, keys never leave your computer, approvals and AI permissions apply.
- MCP tool reference: Every WECOMMANDER MCP tool, grouped: what it does, its risk and approval rule, where it works (agent devices, SSH servers, Linux only) and its key arguments.