PlugWiki
AgentDock logo

AgentDock MCP Server: Let Your Assistant Work on Your Computer or Server

OfficialLocal

Lets Claude, Cursor and other assistants run commands, edit files and keep track of tasks on your own computer or server.

AgentDock is a program you install on a computer, server or container. It then offers file, command and task tools to any assistant that can connect to a web address with a secret token. It is built by its own project, not by an app vendor, and the default address is on your own machine.

There is a Docker image and native installers for macOS, Windows and Linux. You can also pair it with the NexusDock service to reach several devices through one address. This is optional.

It can run any command and change or delete files, and it has no read-only mode. Use the Docker route if you want that kept inside a container.

Good for

  • Developers who want the assistant to work in a real project folder
  • Anyone who manages a VPS or home server and wants to do it by chat
  • People who want one assistant to reach several machines

Works in

12 should work

Tap an app to see how to set it up and any known problems.

Install AgentDock MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the AgentDock MCP server to Claude Code. Run this command:
    
    claude mcp add --transport http agentdock http://127.0.0.1:8765/mcp --header "Authorization: Bearer <AGENTDOCK_AUTH_TOKEN>"
    
    Ask me for AGENTDOCK_AUTH_TOKEN and only put the values in that config, nowhere else.
    
    Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires AgentDock running (Docker or installer), Your AgentDock token.

Read-only mode: There is no read-only option. Run AgentDock in Docker so commands run in a container, not on your own account.

AgentDock must be running before your app can connect. With Docker, download the project's docker-compose.yml, put a random token in a .env file as AGENTDOCK_AUTH_TOKEN, run docker compose up -d and wait until docker compose ps says healthy. The compose file publishes the server on 127.0.0.1 port 8765 by default (AGENTDOCK_PUBLISH_HOST changes this). Native installers for macOS, Windows and Linux are on the releases page. The address and token below are the defaults from the project's docs. The snippet follows the vendor's Connect clients page for this app, with the local address from its Docker page and a token placeholder. Replace the token placeholder, and restart the app so it loads the tools.

From the vendor's docs. Source.

Tools

  • Runs a shell command

    Runs a command on the machine AgentDock is installed on. It waits a few seconds for the result and otherwise returns a running session. You can set a working folder, environment values, a timeout and standard input.

    Try: “Run the test suite in ~/AgentDock/myapp and show me the failures.”

    Delete · exec_command

  • Reads a text file

    Returns a slice of a UTF-8 text file by line range, from any path the host allows. Relative paths start in the AgentDock working folder.

    Try: “Show me lines 1 to 80 of ~/AgentDock/myapp/config.yml.”

    Files · read_file

  • Lists a folder

    Lists folder entries with a depth limit, glob filters and a choice of files or directories only. Hidden and ignored paths are left out unless you ask.

    Try: “List the folders two levels deep under ~/AgentDock.”

    Files · list_dir

  • Searches files for text

    Looks through text files for a word or regular expression and returns the matches. Relative paths search the AgentDock working folder.

    Try: “Find every file that mentions DATABASE_URL.”

    Files · search_text

  • Edits, adds, moves or deletes files

    Changes files through one entry point with five actions: replace text, apply a patch, add a file, delete a file, or move a file.

    Try: “Add a retry to the fetch call in client.py and show me the diff.”

    Delete · file_edit

  • Checks running commands

    Lists the command sessions that exec_command started, or shows one session's status and recent output.

    Try: “Is the build I started still running?”

    Read · session_observe

  • Sends input to or stops a command

    Writes text into a running command session, stops one session, or stops all of them.

    Try: “Stop the dev server I started earlier.”

    Runs code · session_act

  • Shows what AgentDock can do here

    Returns the setup the assistant works in: available capabilities, connected integrations, rules and high-priority context.

    Try: “What can AgentDock do on this machine?”

    Read · agentdock_context

  • Reads project rules

    Reads the AGENTS.md rules that apply to a project folder and the project's local skill index, so the assistant follows them before it works there.

    Try: “Read the project rules for ~/AgentDock/myapp.”

    Files · workspace_context

  • Starts a saved task

    Creates a task record with goals and steps before a long piece of work. Progress is saved so the work can continue after an interruption.

    Try: “Create a task to migrate the database and track each step.”

    Write · task_create

  • Tracks and resumes tasks

    Lists, opens and updates saved tasks. Actions include checkpointing progress, blocking, resuming, final review and completing a task.

    Try: “Resume the migration task where it stopped.”

    Write · task_manage

  • Looks at an image

    Loads an image from a saved artifact, a path on the machine, or a web address and passes it to the assistant as an image.

    Try: “Look at ~/AgentDock/screenshots/error.png and tell me what failed.”

    Files · view_image

Read from the source code

Permissions

high risk · 8 things to watch

High risk

The assistant can run any command and change or delete files as the user that runs AgentDock, and nothing limits it to read-only. Run it in the Docker container, set a token, and keep the address private.

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • It can run any command on the machine

    The command tool runs whatever the assistant asks as the user that runs AgentDock. Run it in the Docker container if you can, and read each command before approving it. github.com

  • It can overwrite and delete files

    The file edit tool can replace, add, move and delete files at any path the host allows, including absolute and ~/ paths. Keep backups and work in a project folder under version control. github.com

  • Without a token, anything local can call it

    On the loopback address the code lets the server start with no token. Always set a token, because any program or page that can reach the port could otherwise run commands. github.com

  • A public address hands over the machine

    The docs describe a Cloudflare tunnel for remote access. Anyone with the address and token can run commands, so keep authentication on and never share the token. docs.nexusdock.co

  • The assistant can add more servers and code

    The add-server tool can register a local command as a new MCP server, and the skill and plugin tools install code that adds tools. Review what the assistant installs. github.com

  • Opt-in browser tools can use your logins

    The README says the browser tools can reuse login sessions. Leave browser tools off, or point them at a separate browser profile. github.com

  • Binds to this machine unless you change it

    The default host is 127.0.0.1, and a non-loopback host will not start without a token or OAuth. The Docker image listens on all addresses inside the container, and the compose file publishes it on 127.0.0.1 by default. github.com

  • Tool text steers the assistant

    Some descriptions tell the assistant to call the context tools first and to create a task before big jobs. This changes how it behaves, so expect extra tool calls. github.com

Security advisories

No published security advisories as of Nov 1, 2026.

Technical details
Turning on read-onlyNo read-only setting exists. The command and file edit tools are always registered, and browser and coding-agent tools are extra when enabled.
Talks toAddresses you ask it to fetch or browse. MCP servers you add with the add-server tool. NexusDock service, only if you pair the device. Cloudflare, only if you turn on a tunnel
Sign-in and keysA bearer token you set (AGENTDOCK_AUTH_TOKEN) or an OAuth sign-in. State, sessions and settings are kept in ~/.agentdock. Pairing with NexusDock keeps the device token in that folder.
Scopes
  • Whatever the user that runs AgentDock can do on the machine

Questions

Is AgentDock an official server?

Yes. AgentDock is the uvwt project's own product, and this server ships in its repo (uvwt/agentdock) with the desktop apps and installers. It connects to your own machine, not to another company's app, so no outside vendor ships a different one.

How do I install AgentDock?

Start AgentDock first, then add its address to your assistant with the install box. The Docker route is the shortest: it needs Docker and takes a few commands. Native installers for macOS, Windows and Linux are on the releases page.

Is AgentDock safe?

Only if you control who can reach it. It can run commands and delete files, and it has no read-only mode. Use a token, keep the address on localhost, and prefer the Docker container.

Does it need an API key?

It needs a token that you create. In the Docker setup you generate a random value and put it in a file named .env. AgentDock also supports an OAuth sign-in, and the desktop installers keep their own connection settings in the control panel.

Can it run read-only?

No. The command and file edit tools are always on, and no setting turns them off.

Do I need NexusDock?

No. AgentDock works alone. Pairing with NexusDock is optional and adds shared notes, workflow templates and one address for several devices, through seven extra tools that are not in the list here.

Which tools are on by default?

19 of the 25 listed. The three browser tools need the browser setting turned on, and the three coding-agent tools need ACP turned on.

Can I use it with ChatGPT?

The README says you can manage computers from ChatGPT in the browser. ChatGPT needs an address it can reach, so the Docker docs describe a temporary Cloudflare tunnel for that. Keep the token on.

Is it free?

The AgentDock server is free and open source under Apache-2.0. Check the NexusDock site for any cost of the paired service.

Information

updated Nov 2, 2026

Publisher
Official
Transport
Streamable HTTP, stdio
Auth
API key
License
Apache-2.0
Last commit
Oct 27, 2026
Open issues
3

WebsiteGitHubChangelogIssues

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 7
  1. github.com · github.com
  2. docs.nexusdock.co · docs.nexusdock.co
  3. docs.nexusdock.co · docs.nexusdock.co
  4. github.com · github.com
  5. github.com · github.com
  6. github.com · github.com
  7. github.com · github.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 2, 2026.