AI access with MCP: how it works
The Model Context Protocol (MCP) is an open standard that lets AI assistants call tools. WECOMMANDER is a remote MCP server: your AI client adds one URL, signs in with your account, and gets tools to read files, run commands and manage servers on the machines of your workspace, with your limits and approvals applied to every call.
One URL for all your machines
https://wecommander.app/mcpEvery AI client that supports remote MCP servers over HTTPS can use it: Claude (web, desktop and Claude Code), ChatGPT, Cursor, VS Code and others. Sign-in uses OAuth with your WECOMMANDER account; clients that only take a bearer token can use an API access key instead.
What happens on a tool call
- The AI calls a tool, for example process_exec with a command and a device.
- The gateway checks the caller's role, the arguments, the device's AI permissions and the safety net.
- A read runs directly. A high-risk action waits for a person's approval of that exact action, unless the device allows it (an allowed command or full trust).
- The gateway sends the action to the device's agent over its outbound encrypted connection. The agent checks it again: allowed folders, its own switches and, for high-risk actions, the signed approval.
- The result comes back with secrets masked, and the call is written to the audit log.
Two ways to connect a server
| Agent | SSH (Hybrid Access) | |
|---|---|---|
| What is installed on the server | The WECOMMANDER agent (Windows, macOS, Linux) | Nothing |
| Path | Server → outbound connection → wecommander.app | wecommander.app → the app on your computer → SSH to the server |
| Works while your computer is off | Yes | No: your computer and the app must be on |
| Folder limit | Allowed folders enforced on the device | None: a command reaches everything the SSH user can |
| Run as root | Linux only, with two keys | Through sudo -n when the computer allows sudo for that server and an admin turns it on |
| Tools | The full set | Commands, files over SFTP, overview, disk, processes, cleanup, process signals, timers and cron, Docker |
| AI permissions | Trust level, allowed commands, safety net | The same, plus read-only sharing that refuses every change |
Choose the agent for always-on or autonomous work and tight folder limits; choose SSH to try quickly or where nothing can be installed. Hybrid Access is available from the Windows, macOS and Linux apps.
What the AI sees
devices_list lists agent devices; targets_list lists every place the AI can act on (agent devices and shared SSH servers) with each one's trust level, run as, safety net and supported tools, so the AI knows whether it is root and whether an action will need approval.
Limits that always apply
- AI clients cannot change their own permissions, trust levels, allowed commands or autonomous mode; those need an admin in a browser session.
- Approvals are single-use and bound to the exact action and device.
- On the Free plan, a monthly number of tool calls is included; when it is used up the AI is told when it resets and where to upgrade.
- Connect Claude (web, desktop and Claude Code): Add WECOMMANDER to Claude as a custom connector with https://wecommander.app/mcp and OAuth, or to Claude Code from the command line.
- Connect ChatGPT: Add WECOMMANDER to ChatGPT as a custom MCP connector with https://wecommander.app/mcp and OAuth sign-in.
- Connect Cursor, VS Code (Copilot) and other MCP clients: Use https://wecommander.app/mcp from Cursor, VS Code with GitHub Copilot and any MCP client: OAuth sign-in, or an API access key as a bearer token.
- 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.
- 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.