uvwt/agentdock

Secure MCP runtime for AI agents to operate local machines, servers, and containers with multi-device orchestration.

What it solves

AgentDock provides a secure, controlled way for AI agents (like ChatGPT or Claude) to operate real-world environments. It eliminates the need to manually switch between SSH sessions or use limited coding-agent quotas by allowing agents to manage multiple local computers, remote servers, and containers from a single conversation.

How it works

It acts as an independent tool runtime that exposes capabilities via the Model Context Protocol (MCP). AI agents connect to AgentDock instances running on various machines; the agent then sends commands to these instances to perform tasks. It supports multiple connection modes, including local-only, temporary public addresses, and fixed domains via Cloudflare Tunnels, with authentication handled via Bearer Tokens or OAuth.

Who it’s for

Developers and system administrators who want to automate the management of their infrastructure, deploy code, and troubleshoot servers using AI agents without leaving their chat interface.

Highlights

  • Multi-device coordination: Manage multiple hosts (Mac, Linux, Windows, VPS) in one conversation.
  • Comprehensive toolset: Includes file system access, shell command execution, Git operations, and browser automation.
  • Extensibility: Supports "Skills" and dynamic MCP servers to add new capabilities.
  • Secure boundaries: Implements permission boundaries, authentication, and structured, traceable results.
  • Task persistence: Ability to persist long-running task states and resume after interruptions.
  • Browser/Desktop automation: Native control of Chrome/Chromium and macOS desktop applications.

Related

  • Project
    asheshgoplani/agent-deck

    A command center for AI coding agents that allows users to manage, monitor, and orchestrate multiple agent sessions across different projects from a single terminal.

  • Project
    crabtalk/crabtalk

    An agent daemon that runs AI agents, dispatches tools, and connects to MCP servers, using restricted harnesses for secure tool execution.

  • Project
    bcurts/agentchattr

    A local chat server that enables real-time coordination between humans and multiple AI coding agents, eliminating the need to copy-paste prompts between terminals.

  • Project
    barry-ran/QuickDesk

    An AI-native remote desktop tool with a built-in MCP server that enables AI agents to view and control remote computers globally.

  • Project
    pocketpaw/pocketpaw

    A self-hosted AI agent that runs locally on your machine, providing a native desktop app and integration with multiple messaging platforms like Discord and Slack.