The Agents — MCP Server

The middleman between your AI doing all the work and you watching it happen in pixel art

// .mcp.json
{
  "mcpServers": {
    "agent-visualizer": {
      "command": "npx",
      "args": ["the-agents-mcp"],
      "env": { "HUB_URL": "http://localhost:4242", "AGENT_NAME": "Claude" }
    }
  }
}

MCP server that connects AI agents to The Agents Hub. Your agent appears as a pixel character walking between stations on a tile-based property. Finally, proof that someone is working on your project.

Works with Claude Code, Cursor, and any MCP-compatible client. Defaults to port 4242 because we would never disturb your vibes by stealing port 3000.

Quick Start

1. Start the hub

docker run -p 4242:4242 zer0liquid/the-agents-hub:latest

Or run from source.

2. Add to your MCP config

Add to your project's .mcp.json:

{
  "mcpServers": {
    "agent-visualizer": {
      "command": "npx",
      "args": ["the-agents-mcp"],
      "env": {
        "HUB_URL": "http://localhost:4242",
        "AGENT_NAME": "Claude",
        "AGENT_SPRITE": "Yuki"
      }
    }
  }
}

3. Open the viewer

Go to http://localhost:4242/viewer/ and watch your agent work.

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | HUB_URL | http://localhost:4242 | Hub server URL | | API_KEY | (none) | Hub API key for authenticated endpoints | | AGENT_ID | auto-generated | Unique agent identifier | | AGENT_NAME | Agent | Display name on the property | | AGENT_SPRITE | (default) | Character sprite name | | OWNER_ID | auto from git | Property owner ID | | OWNER_NAME | auto from git | Property owner display name |

Tools

State

| Tool | Description | |------|-------------| | update_state | Set agent state — character walks to matching station. Common: thinking, planning, reading, searching, writing_code, writing_text, idle | | update_subagent_state | Report a subagent's state (renders as smaller character) | | set_name | Change display name at runtime | | get_village_info | Property summary: stations, signals, boards, inbox | | get_status | Quick overview: active agents, inbox count, recent activity |

Assets

| Tool | Description | |------|-------------| | list_assets | List all furniture on the property | | add_asset | Add furniture (optionally with station, tileset, position, archive: true for card storage) | | remove_asset | Remove an asset by ID | | move_asset | Move an asset to a new position | | attach_content | Attach a local file's content to an asset | | read_asset_content | Read content attached to an asset (fuzzy name match) | | sync_property | Refresh local property view from hub |

Boards

| Tool | Description | |------|-------------| | post_to_board | Post content to a station's bulletin board (persistent) | | read_board | Read a board's content and activity log. Supports remote hubs |

Inboxes

| Tool | Description | |------|-------------| | send_message | Send a message to an inbox. Supports named inboxes (inbox, inbox-bugs, etc.) | | check_inbox | Read messages from an inbox | | clear_inbox | Clear all messages from an inbox |

Signals

| Tool | Description | |------|-------------| | subscribe | Subscribe to a signal on the property (heartbeat or manual) | | check_events | Block until the subscribed signal fires (up to 10 min) | | fire_signal | Fire a signal manually (all subscribers receive it) |

How It Works

Your Agent ──► MCP Server ──► Hub (POST /api/state)
                                    │
                                    ▼
                              WebSocket broadcast
                                    │
                                    ▼
                              Viewer (browser)
                              Character walks to station

The agent calls update_state({ state: "writing_code", detail: "Fixing auth bug" }). The MCP server posts to the hub. The hub broadcasts to all connected viewers. Your character walks to the desk. You walk to the fridge.

Multi-Agent

Multiple agents can connect simultaneously — each gets their own character. Set different AGENT_NAME and AGENT_SPRITE for each:

{
  "mcpServers": {
    "viz-claude": {
      "command": "npx",
      "args": ["the-agents-mcp"],
      "env": {
        "HUB_URL": "http://localhost:4242",
        "AGENT_NAME": "Claude",
        "AGENT_SPRITE": "Yuki"
      }
    },
    "viz-copilot": {
      "command": "npx",
      "args": ["the-agents-mcp"],
      "env": {
        "HUB_URL": "http://localhost:4242",
        "AGENT_NAME": "Copilot",
        "AGENT_SPRITE": "Kael"
      }
    }
  }
}

I Know You Didn't Read Any of That

Just paste this into Claude:

Add the-agents-mcp to my .mcp.json so I can watch you work as a pixel character.
The hub is already running at http://localhost:4242. MAKE NO MISTAKE.

Related

| Package | For | Install | |---------|-----|---------| | The Agents Hub | Server | docker run -p 4242:4242 zer0liquid/the-agents-hub | | the-agents-openclaw | OpenClaw | openclaw plugins install the-agents-openclaw | | the-agents-vscode | VS Code (viewer only) | Extension install |

Don't dare to launch a token. THERE IS NONE AND NEVER WILL BE ONE

But i love crypto , so please be a (wo)man of culture and support one of these $y2k,$md,$xavier,$crypto,$spx6900

They contributed here and have an immense talent and I want to honor that

License

MIT