Agent Guild
Runs Docker Agent (docker agent) sessions in real terminals, including installing and updating the downloaded Docker CLI plugin and reporting which Docker CLI copy is used.
Discovers and manages tmux sessions alongside herdr sessions, showing them as cards with helper agents and model information so MCP clients can see and steer them.
Run Claude Code, Codex CLI, Antigravity CLI, Grok Build, Docker Agent and your own shells side by side in real terminals. Pick the same live sessions up from your phone, VS Code or any browser. Free, on Windows, macOS and Linux.
Agent Guild is a small manager that runs on your computer and serves one local web page. Close the page whenever you like; the sessions keep running.
Try the demo: walk the Yard in your browser. Simulated, free, no account and nothing installed.
Quick start
Or install with npm, on Node.js 22 or newer:
npm install -g @oddessentials/agent-guild
agent-guildThe page opens in your browser. Press New on a tool's card to start a session. Sign in to each tool inside its first session, as you would in any terminal. To look around first, try the demo (simulated, nothing installed) or watch the trailer.
Or ask your agent. Point it at
the setup skill,
or run claude plugin marketplace add oddessentials/agent-guild --sparse .claude-plugin plugins
and claude plugin install agent-guild@agent-guild. It installs and checks Agent Guild,
then connects over MCP to see and steer your sessions.
Related MCP server: cmuxlayer
Why Agent Guild
The real CLIs, unchanged. Each session is the tool itself in a real terminal, not a chat copy of it. Whatever sign-in the tool uses works: a subscription, an API key, Amazon Bedrock or your own gateway.
Every agent in one place. Claude Code, Codex CLI, Antigravity CLI, Grok Build, Docker Agent, any shell, and tmux or herdr sessions, each card showing its helper agents and model, with usage meters for Claude Code and Codex CLI.
The same session on every screen. Any browser, a phone over Tailscale with touch keys, VS Code, and the Agent Guild app: iPhone, iPad and Apple Vision Pro on the App Store; Android, Android TV, Google TV and Android XR on Google Play or GitHub; Fire tablets and Fire TV on the Amazon Appstore; and Meta Quest on the Meta Horizon Store. Apple Watch and Meta Ray-Ban Display list your sessions and read a session's last answer aloud. See what each screen needs.
Native on Windows. No WSL. Also macOS, Linux, and headless Linux servers over SSH.
Nothing in between. No account and no relay server: the manager listens on
127.0.0.1, and other devices reach it through your own Tailscale network or reverse proxy. See Privacy.
How it compares
Agent Guild | tmux over SSH | Claude Code Remote Control | |
Agents | Any CLI, side by side | Any CLI | Claude Code |
On a phone | The real terminal in a web page or app, with touch keys | An SSH app | A chat view in the Claude app |
Sign-in | Whatever the tool uses | Whatever the tool uses | Pro, Max, Team or Enterprise; not API keys, Bedrock or a gateway |
Windows without WSL | Yes | No | Yes |
Not yet: push notifications. Phones and the Apple apps reconnect when you open them again; the sessions keep running on the computer in the meantime.
License: open source under the GNU AGPL 3.0. Free for personal and commercial use.
Requirements
Node.js | 22 or newer |
Windows | 10 version 1809 or later, x64 or arm64. No WSL needed. |
macOS | 11 or later, Intel or Apple silicon |
Linux | x64 or arm64 with glibc 2.28 or later (for example Debian 10, Ubuntu 20.04 or RHEL 8), with a desktop or over SSH |
No compiler is needed, except on Linux without glibc, such as Alpine:
agent-guild prints the commands that build its terminal library there. A
missing tool shows Install on its card; Antigravity CLI, which is not on
npm, shows its install command instead. For Docker Agent, see
Docker Agent.
Agent Guild for VS Code opens the same page in an editor tab. Editors that install from Open VSX get it there.
Setup
Access token
You do not create one. The first run generates it, and agent-guild opens
the page with it. If a browser asks for it, open the link that
agent-guild url prints. Anyone with that link can use your terminals.
To replace it, stop the manager (this ends running sessions), delete
auth-token from the data folder, and
start again.
Docker Agent
Docker Agent is the docker agent plugin; Docker Desktop 4.63 and later ship
it, and its card's Install downloads the latest release into
~/.docker/cli-plugins when no copy runs. The card's installation box lists
every copy the Docker CLI finds and which one runs; the card updates and
removes only the downloaded one, and leaves Docker Desktop's to Desktop.
On Linux, Docker Engine does not include the plugin; the card's Install
adds it. The card needs Docker's own docker command: Podman's
docker and the Ubuntu snap of Docker do not load the plugin. If you
installed Docker Agent on its own as docker-agent (Homebrew or a release
download), add this to providers.json,
then run agent-guild restart:
{ "providers": [{ "id": "docker", "command": "docker-agent", "args": ["run"], "versionArgs": ["version"] }] }Headless Linux server
On the server:
npm install -g @oddessentials/agent-guild agent-guild --no-browserIt prints
Open: http://127.0.0.1:47821/#token=….On your computer, open a tunnel and leave it open:
ssh -L 47821:127.0.0.1:47821 you@serverOpen the printed link in your browser. The local port must be
47821; any other port is refused.
To start Agent Guild when the server boots, choose Settings → Startup → When the computer starts in the page, then run once on the server:
sudo loginctl enable-linger $USERThis needs systemd and is not offered under WSL. agent-guild status shows
the result.
Remote access with Tailscale
Use your terminals from a phone or another computer on your Tailscale network.
You need Tailscale 1.52 or later, signed in on both devices in the same tailnet, with MagicDNS on.
In the page, choose Settings → Remote access → Enable remote access.
If it shows Open Tailscale approval, approve HTTPS there, then choose Continue setup.
If it says Tailscale needs permission, run the command under Connection details in a terminal allowed to manage Tailscale, then choose Continue setup.
Choose Connect another device and scan the QR code, or open the sign-in link, on the other device. On a TV or another device without a camera, enter your remote address in the Agent Guild app and approve the code it shows.
The sign-in link contains your access token. Disable remote access turns it off; sessions keep running. On a headless server, do this through the SSH tunnel above. For another reverse proxy, see configuration.
macOS keychain
The first Claude Code usage meter may ask for access to the "Claude Code-credentials" keychain item. Choose Always Allow.
Features
Real terminals that survive closing or reloading the page
Two terminals side by side with Split, a text size control, and Copy… to select terminal text
Touch keys for phones and tablets: arrows, Tab, Enter, Esc, Ctrl+C, Paste and Clear
Dictate into a terminal and Read the agent's last answer aloud (Settings → Voice and Read aloud)
A working-folder picker with recent folders and a folder browser, and a button that opens the folder in your file manager
Install and update each tool from its card, and upgrade Agent Guild from the page
Helper agents and the model in use, shown on each card (setup per tool)
Rate-limit meters for Claude Code and Codex CLI
Several accounts per tool (accounts)
Model grades and benchmarks for each tool's models
Resume a tool's earlier sessions with Existing…
Read Google's saved conversations with History, filtered to the working folder
See what Claude Code, Codex CLI and Grok Build remember with Memory, and which instruction files they and Antigravity CLI read in the working folder with Instructions
GitHub issues, pull requests, workflow runs, branches and cloning
The manager's environment on the Shell card: runtimes, design tools, Docker, and the versions a project pins
Start the manager when you sign in, on Windows, macOS and Linux
Shared notes, a news feed, skins, light and dark mode
Alert sounds when the manager stops or an update is found
Videos: Agent Guild's trailers and tutorials
The Yard: your tools and sessions as a 3D world, one for each skin
tmux and herdr sessions that keep running when the manager stops
The Yard
For fun, the Yard shows your tools and sessions as a 3D world, one for each skin. A double-click on a character opens its real terminal. The clip at the top of this page shows it, and the demo lets you try every skin.
Commands
Command | What it does |
| Start the manager if needed and open the page. |
| Show whether the manager is running and list its sessions. |
| Stop the manager and end every session. |
| Restart the manager on the installed version and end every session. |
| Print the page link with the access token. |
| Launch fireworks over every open Yard. A coding agent in any terminal can run it too. |
| Run the manager in the foreground. |
| Diagnose the health of the installation and related tooling. |
| Serve MCP on stdio, so a coding agent can see and steer the sessions. See Agent Guild for agents. |
| Print the installed Agent Guild version ( |
Sessions end when the manager stops or the computer restarts. tmux and herdr sessions survive a manager stop.
After upgrading Agent Guild and restarting the manager, refresh every open Agent Guild browser page to load the new frontend. Pages reconnect automatically, but keep their loaded JavaScript until refreshed. Refreshing is required for compatibility with the upgraded manager.
Privacy
The manager listens on 127.0.0.1 only. Remote access goes through a private
Tailscale route or your own reverse proxy to that address. None of these
requests carry your code or prompts:
To | For | How often |
npm registry, Homebrew, GitHub releases (Docker Agent) | Version checks | Hourly |
Anthropic and OpenAI | Usage meters, with the tool's own sign-in | Every minute while the page is open |
OpenRouter | Model list for benchmarks | Every 6 hours |
News feeds, Hacker News, arXiv, GitHub | News feed | Every 30 minutes while the page is open |
GitHub releases | What's new | Hourly, more often just after a release |
herdr.dev, Homebrew | Multiplexer version checks | Hourly |
GitHub, GitHub Status | GitHub panel | When you use it, and every few seconds while a workflow run you chose to watch is going |
Package sources | Installs and updates | When you ask |
YouTube | Video thumbnails and playback | When you open Videos |
Failed news, model list and release checks retry after 10 minutes.
AGENT_GUILD_NO_UPDATE_CHECK=1 turns off version checks and self-upgrade.
More
Agent Guild for agents: the MCP server, and a plugin and skill that set Agent Guild up
Configuration: add tools, accounts, environment variables, reverse proxies
Agent reporting: show agents from any tool
Skins: make your own look
Agent Guild: World
Agent Guild: World is a free, living 3D world for your AI coding agents, coming to Steam in November 2026: six worlds to roam with a free camera, your own characters and companions from Agent Guild: Studio or the Steam Workshop, a Guild Board for your notes and to-dos, and controller support, on Windows, macOS and Linux. Wishlist it on Steam.
Follow
Support
Agent Guild is free to use. If it saves you time, buy the lab a beer. It is a tip, not a tax-deductible donation, and it changes nothing about the software.
License
Copyright (c) 2026 Odd Essentials. Licensed under the GNU Affero General Public License 3.0.
Available Tools
8 toolsfireworksFireworks over the YardA
Launch fireworks over the Yard on every screen connected to Agent Guild. Use it to celebrate something the person cares about, such as tests, a build or a release that passed. Says how many pages saw them.
Inputs: no named parameters.
guild_statusAgent Guild statusARead-only
Whether Agent Guild is running on this computer, its version, and how many sessions it has. To customize a Yard, the characters and companions of a Yard or Agent Guild: World, or to install Grove Companion on the desktop, follow https://agentguild.oddessentials.ai/agents/.
Inputs: no named parameters.
list_sessionsList sessionsBRead-only
Every session: status, recent activity (working), model, sub-agents, latest reply id, history cursor, and inputInstance for keyed input.
Inputs: no named parameters.
read_repliesRead reply historyARead-only
Read retained replies in order, with ids and cursors. Pass your last handled cursor as after; omit it to start at the oldest retained reply. Reads do not consume replies. Check gap for evicted or reset history and save each cursor only after handling its reply.
Inputs: session_id (required), after, limit.
read_replyRead a replyARead-only
The whole text of the answer that ended a session's last turn, as Markdown. Only tools with reporting on (Claude Code, Codex CLI, Antigravity CLI, Grok Build, Docker Agent) have one.
Inputs: session_id (required).
read_screenRead a terminalARead-only
The text a session's terminal shows right now, as plain lines. Use it to see a question the tool is waiting on.
Inputs: session_id (required).
start_sessionStart a sessionA
Start a new session of a coding tool or shell in Agent Guild. Ask the person first. It opens on every screen connected to Agent Guild.
Inputs: tool (required), cwd, name, resume, +1 more.
stop_sessionStop a sessionADestructiveIdempotent
End a session's process. Ask the person first. The session stays listed, with its final screen, until it is removed in Agent Guild.
Inputs: session_id (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 8 tools
Most tools target distinct actions: status, listing, reading, starting, stopping, and celebrating. The main possible confusion is between read_reply (last turn's final answer) and read_replies (cursor-based batch of retained replies), but the descriptions make the distinction workable. guild_status versus list_sessions also overlaps slightly on session information but remains separable.
The set is mostly verb_noun (list_sessions, read_reply, read_screen, start_session, stop_session), which is readable. However, guild_status and fireworks break the verb-first pattern, and read_reply versus read_replies introduces singular/plural naming that could be clearer.
Eight tools is a well-scoped count for a session management and monitoring server. Each tool appears to have a distinct role, with no obvious redundancy or bloat.
The surface covers status, session listing, reading outputs, screen inspection, and session start/stop. But there is no tool to send keyed input or answer a prompt, even though read_screen is explicitly meant to see questions the session is waiting on; that is a significant dead end. There is also no remove_session tool, though stop_session mentions removal happens elsewhere.
Maintenance
Related MCP Connectors
Real-time chat for AI agents. Claude Code, Cursor, Cline and Codex join channels over MCP.
Remote MCP server exposing your Node Command fleet (files, shell, processes, browser) to AI hosts.
Shared memory MCP server for Claude Code, Codex, Gemini CLI and Cursor. Cited, editable.
StremAI MCP: shared memory for AI coding agents. Connected agents can recall. OAuth + local stdio.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceTerminal MCP server for AI coding agents with persistent PTY sessions, ring-buffer incremental reads, headless xterm screen capture, multi-agent orchestration, and a real-time web dashboard.38 npm28MIT
- AlicenseAqualityBmaintenanceTerminal multiplexer MCP server for orchestrating parallel AI agents. Manages workspaces, panes, surfaces with send_input/read_screen/spawn_agent/stop_agent tools. Supports Claude Code, Codex, Gemini, Cursor CLI agents with lifecycle management, browser automation, and agent status push via Claude --channels.1030Apache 2.0
- FlicenseNot gradedqualityBmaintenanceThe terminal AI agents can drive: a cross-platform desktop terminal (macOS/Linux/Windows, MIT) that runs a local MCP server. Spawn tabs/panes, run commands with structured output, read screens and full scrollback, take scrolling screenshots, record sessions with secret redaction, switch identity profiles. Auto-registers with Claude Code / Codex / Gemini CLI on install.18-
- AlicenseNot gradedqualityDmaintenanceLocal multi-harness orchestration for Claude Code, Grok CLI, and AGY CLI, enabling spawning, management, and coordination of AI coding agents in a visible terminal grid.2MIT