MCP integration

Cmdop’s MCP integration works in two directions. Cmdop’s own MCP server is a fleet interface for a local assistant such as Claude Code or Codex: it operates joined machines through Cmdop, and it is not a general local filesystem or shell interface. In the other direction, a Cmdop agent can call tools published by an outside MCP server once you have reviewed and approved them.

Both are built on the Model Context Protocol . Pick the one that matches what you are trying to do.

Let your assistant work across your fleet

You already use Claude Code or Codex on your own computer, and you want it to reach the machines you joined to Cmdop.


cmdop mcp install --agent claude
cmdop mcp status
cmdop mcp tools

The assistant gets a fleet interface: list machines, ask a machine’s agent to do something, run one command, read a file or list a directory, and read Machine Care facts — plus higher-level actions such as driving a machine’s browser, web search, and publishing a site. It uses this machine’s existing Cmdop access, so there is no second account to create and no local endpoint to open. It is not a copy of your local terminal, editor, or filesystem. See what the assistant can actually do.

Start here: Claude Code and Codex.

Let a Cmdop agent use someone else’s tools

You want a Cmdop agent to call tools published by another MCP server — a browser driver, a code-hosting integration, an internal service.

Cmdop saves the server definition disabled. Before an agent can call anything, you grant consent to the exact command, run one contained test, and approve the exact tool catalogue you reviewed. An approved server starts on its first real tool call, not when the machine boots.


cmdop mcp servers list

Start here: External MCP tools.

Common questions

Can an MCP client run commands on a remote machine through Cmdop?

Yes. Cmdop’s fleet MCP server gives a local assistant tools to list joined machines, ask a machine’s agent to do work, and run bounded commands on a named machine. The target still has to be online and may require its connection PIN; registration alone is not reachability.

Does Cmdop expose my local shell to Claude Code or Codex?

No. Claude Code and Codex already have their own local shell and file tools for the machine they run on. Cmdop’s fleet MCP server adds a fleet interface to other joined machines; it is not a duplicate of the assistant’s local terminal or filesystem.

Can a Cmdop agent use external MCP tools safely?

Only after review. Cmdop saves external MCP server definitions disabled, then requires review, a contained test, and approval of the exact tool catalogue. If the external server is reachable over HTTP or SSE, secure that network surface before approval — see MCP troubleshooting.

Discovery is not installation

cmdop mcp discover reads MCP definitions that other local clients already have, without starting them and without trusting them. A definition Cmdop found is not a definition Cmdop owns, and its runtime state is unknown until you import and approve it.

The public marketplace works the same way. A catalogue listing tells you a capability exists; it never tells you the capability is installed, approved, or available on a machine.