This is an early release preview. You may encounter bugs.
AgentDock logo
Unclaimed

Tool interface security

AgentDock

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

A 82/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
Discussion

01 / Summary

About AgentDock.

AgentDock is a tool runtime that lets AI agents operate real machines within explicit permission boundaries. Installed on a local computer, a remote server, or a container, it exposes file, command, Git, Skill, MCP, browser automation, and task execution tools over MCP, and returns structured, traceable results. It runs no chat interface and no model inference of its own: a client such as ChatGPT, Claude, or Codex connects to it.

Several AgentDock instances can connect to one client at once, so a single conversation can coordinate work across a laptop, a LAN host, and a cloud VPS: writing code and running tests in the real environment, managing Docker services, reverse proxies, and deployments, or inspecting logs, processes, and runtime state. Remote access runs locally only, through a temporary public tunnel address, or through a fixed domain on a Cloudflare Tunnel, with bearer-token or OAuth authentication.

Tasks persist their goals, steps, checkpoints, and blockers so long-running work resumes after an interruption. AgentDock can also act as a native Agent Client Protocol (ACP) client that hosts a local coding agent, with presets for Codex, Claude, and Grok, and can pair with NexusDock for long-term project memory, runbooks, workflow templates, and multi-device state.

  • Files and commands: text search, structured edits, atomic writes, path boundaries, and command execution with timeouts, PTY sessions, and output redaction
  • Multi-device orchestration: connect several instances to one client and coordinate work across machines in one conversation
  • Skills and dynamic MCP: install and update managed Skills, and register, enable, and call dynamic MCP servers over Streamable HTTP or stdio
  • Native ACP: hosts a local coding-agent adapter with presets for Codex, Claude, and Grok, and relays its permission requests
  • Browser and desktop automation: browser sessions with persistent logins, page inspection, screenshots, system Chrome, and macOS desktop control
  • Recoverable tasks: persisted goals, steps, checkpoints, and blockers, with evidence-based completion checks and reusable templates
  • Remote access options: local only, a temporary tunnel address, or a fixed Cloudflare-managed domain, with bearer-token or OAuth authentication
  • NexusDock integration: optional long-term project memory, runbooks, private notes, and multi-device state coordination
  • Cross-platform: the same tool model on macOS, Linux, Windows, WSL, and containers

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

04 / Build

Build with AgentDock.

Browse the catalog for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get AgentDock →

Browse the catalog