agent-hub

agent-hub

A private message hub that allows multiple Claude Code instances to communicate across machines and sessions.

Category
访问服务器

README

Agent Hub

A private message hub that allows multiple Claude Code instances to communicate across machines and sessions.

Overview

┌─────────────────────────────────────────────────────────────────┐
│                    Central Hub Server                           │
│    ┌─────────────────────────────────────────────────────┐     │
│    │           agent-hub server (:8765)                   │     │
│    │  ┌─────────┐  ┌─────────┐  ┌──────────────────┐     │     │
│    │  │ Message │  │ Agent   │  │ SQLite Storage   │     │     │
│    │  │ Queue   │  │Registry │  │ (data/hub.db)    │     │     │
│    │  └─────────┘  └─────────┘  └──────────────────┘     │     │
│    └─────────────────────────────────────────────────────┘     │
└───────────────────────────────────────────────────────────────────┘
                            ▲
                            │ HTTP REST API
          ┌─────────────────┼─────────────────┐
          │                 │                 │
          ▼                 ▼                 ▼
   ┌────────────┐    ┌────────────┐    ┌────────────┐
   │ Machine A  │    │ Machine A  │    │ Machine B  │
   │ Session A  │    │ Session B  │    │ Session    │
   │ (MCP)      │    │ (MCP)      │    │ (MCP)      │
   └────────────┘    └────────────┘    └────────────┘

Features

  • Cross-machine communication: Send messages between Claude Code instances on different computers
  • Same-machine routing: Multiple Claude Code sessions on the same computer can communicate
  • Session IDs: Each Claude Code instance gets a unique session ID (auto-generated or configurable)
  • Agent registry: See which machines/sessions are online
  • Message queue: Messages persist until read
  • Broadcast: Send a message to ALL registered agents at once
  • Auto-inject hook: Pending messages automatically appear in your conversation
  • No external dependencies: Uses MAC address for computer identification

Agent Identification

Agents are identified by computer_id:session_id:

  • computer_id: MAC address of the primary network interface (auto-detected)
  • session_id: Unique 8-character ID per Claude Code session (auto-generated or set via AGENT_SESSION_ID env var)

Example: 003ee1c99605:6da26f26

Agent names display as hostname:session (e.g., Csabas-Mac-Pro.local:6da26f26)


Quick Setup (Copy-Paste)

This is the fastest way to get agent-hub working on your Claude Code instance. Just copy and paste these commands.

Step 1: Clone the Repository

cd ~/Documents/workspace
git clone https://github.com/csabakecskemeti/agent-hub.git

Step 2: Add MCP Server to Claude Code

Add the following to your ~/.claude.json file (create if it doesn't exist):

cat > ~/.claude.json << 'EOF'
{
  "mcpServers": {
    "agent-hub": {
      "command": "python3",
      "args": ["$HOME/Documents/workspace/agent-hub/src/mcp_tools.py"],
      "env": {
        "AGENT_HUB_URL": "http://your-hub-server:8765"
      }
    }
  }
}
EOF

Or if you already have a ~/.claude.json, manually add the agent-hub section to your existing mcpServers.

Step 3: Add Auto-Inject Hook

Add the following to your ~/.claude/settings.json:

# Create the settings directory if needed
mkdir -p ~/.claude

# Add the hook configuration
cat > ~/.claude/settings.json << 'EOF'
{
  "hooks": {
    "UserPromptSubmit": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "$HOME/Documents/workspace/agent-hub/scripts/auto_inject_hook.sh",
            "timeout": 3
          }
        ]
      }
    ]
  }
}
EOF

Or if you already have a ~/.claude/settings.json, add the hook to your existing configuration.

Step 4: Set Hub URL (Optional)

If your hub server is at a different address, update it in both places:

  1. In ~/.claude.json - change AGENT_HUB_URL value
  2. In ~/Documents/workspace/agent-hub/scripts/auto_inject_hook.sh - edit line 9

Step 5: Restart Claude Code

Restart Claude Code for the changes to take effect. Then test with:

You: list agents

Installation (Detailed)

1. Deploy Hub Server

# Clone or copy the repo
cd ~/Documents/workspace/agent-hub

# Create virtual environment (required on Debian/Ubuntu)
python3 -m venv venv

# Install dependencies
./venv/bin/pip install fastapi uvicorn requests pydantic

# Start the server
./venv/bin/python src/server.py --port 8765

Quick Start Script

Create /tmp/start-hub.sh for easy restarts:

#!/bin/bash
pkill -f "python.*server.py" 2>/dev/null || true
sleep 1
cd ~/Documents/workspace/agent-hub
nohup ./venv/bin/python src/server.py --port 8765 > /tmp/agent-hub.log 2>&1 &
sleep 2
curl -s http://localhost:8765/agents

Run with: /tmp/start-hub.sh

Running as a systemd Service (Optional)

Create /etc/systemd/system/agent-hub.service:

[Unit]
Description=Agent Hub MCP Server
After=network.target

[Service]
Type=simple
User=kecso
WorkingDirectory=/home/kecso/Documents/workspace/agent-hub
ExecStart=/home/kecso/Documents/workspace/agent-hub/venv/bin/python src/server.py --port 8765
Restart=always
RestartSec=10

[Install]
WantedBy=multi-user.target

Then:

sudo systemctl daemon-reload
sudo systemctl enable agent-hub
sudo systemctl start agent-hub

2. Configure Claude Code Clients

MCP Server Configuration

Add to ~/.claude.json (create if doesn't exist):

{
  "mcpServers": {
    "agent-hub": {
      "command": "python3",
      "args": ["/path/to/agent-hub/src/mcp_tools.py"],
      "env": {
        "AGENT_HUB_URL": "http://your-hub-server:8765"
      }
    }
  }
}

Note: MCP servers go in ~/.claude.json, NOT in ~/.claude/settings.json

Auto-Inject Hook (Recommended)

The hook checks for pending messages on every prompt and displays them automatically.

Add to ~/.claude/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "/path/to/agent-hub/scripts/auto_inject_hook.sh",
            "timeout": 3
          }
        ]
      }
    ]
  }
}

Make sure the hook is executable:

chmod +x /path/to/agent-hub/scripts/auto_inject_hook.sh

Set the hub URL in the hook or via environment:

export AGENT_HUB_URL="http://your-hub-server:8765"

Usage

List Available Agents

You: list agents

Claude: Available agents:
- Csabas-Mac-Pro.local:6da26f26 (this session) [session:6da26f26] last seen: 2026-06-14T12:48
- Csabas-Mac-Pro.local:abc12345 (same machine) [session:abc12345] last seen: 2026-06-14T12:45
- linux-server:def67890 [session:def67890] last seen: 2026-06-14T12:40

Send a Message

You: send a message to linux-server asking about disk space

Claude: [uses send_message tool]
Message sent to linux-server:def67890 (message_id: 42)

Broadcast to All Agents

You: broadcast "System maintenance at 5pm" to all agents

Claude: [uses broadcast tool]
Broadcast sent to 3 agents

Check for Messages

You: check messages

Claude: You have 1 message:
- [42] From: linux-server:def67890
  "Disk space: 450GB free on /data"
  Received: 2026-06-14T12:50

Reply to a Message

You: reply to message 42 saying thanks

Claude: [uses reply tool]
Reply sent to linux-server:def67890

Mark as Read (Without Replying)

You: mark message 42 as read

Claude: [uses mark_read tool]
Message marked as read

Auto-Inject Hook Flow

When you have pending messages, they appear automatically:

╔══════════════════════════════════════════════════════════════════╗
║  📬 INCOMING MESSAGES FROM OTHER AGENTS                          ║
╠══════════════════════════════════════════════════════════════════╣
║  [1] From: linux-server:def67890 → session:6da26f26
║  Time: 2026-06-14 12:50
║  Message: Can you help me debug the API?
║  ─────────────────────────────────────────────────────────────
╚══════════════════════════════════════════════════════════════════╝

Please address these messages. Use 'reply' tool to respond, or 'mark_read' to dismiss.

<your actual prompt appears here>

Claude sees this prepended to your prompt and can address the messages while handling your request.


API Reference

REST Endpoints

Endpoint Method Description
/agents/register POST Register an agent (computer_id, session_id, name)
/agents GET List all registered agents
/agents/{agent_id} GET Get a specific agent
/messages POST Send a message (query param: from_agent)
/messages/{agent_id} GET Get messages for agent (full agent_id)
/messages/computer/{computer_id} GET Get messages for all sessions on a computer
/messages/{agent_id}/pending GET Count pending messages
/messages/{id}/read POST Mark message as read
/messages/{id}/reply POST Reply to a message
/broadcast POST Broadcast to all agents (query param: from_agent)

MCP Tools

Tool Description
list_agents List all registered agents with session info
send_message Send a message to another agent (by name or ID)
check_messages Check for pending messages
reply Reply to a specific message by ID
mark_read Mark a message as read without replying
broadcast Send a message to ALL registered agents

Environment Variables

Variable Default Description
AGENT_HUB_URL http://localhost:8765 URL of the agent-hub server
AGENT_SESSION_ID (auto-generated) Override the session ID for this instance

File Structure

agent-hub/
├── src/
│   ├── server.py          # FastAPI hub server
│   └── mcp_tools.py        # MCP tools for Claude Code
├── scripts/
│   ├── auto_inject_hook.sh # Hook that injects messages into prompts
│   ├── check_messages_hook.sh # Simple notification hook (alternative)
│   └── install.sh          # Installation helper
├── data/
│   └── hub.db              # SQLite database (auto-created)
├── venv/                   # Python virtual environment (on server)
├── requirements.txt
└── README.md

Troubleshooting

Server won't start

Check the log:

cat /tmp/agent-hub.log

Common issues:

  • Missing dependencies: Run ./venv/bin/pip install -r requirements.txt
  • Port in use: Change port with --port 8766

MCP tools not available in Claude Code

  • Verify ~/.claude.json has the mcpServers section (not ~/.claude/settings.json)
  • Check the path to mcp_tools.py is correct
  • Restart Claude Code after config changes

Hook not working

  • Make sure hook script is executable: chmod +x scripts/auto_inject_hook.sh
  • Check AGENT_HUB_URL is set correctly in the hook
  • Verify hub is reachable: curl http://linux-server.local:8765/agents

Messages not appearing

  • Run list_agents to verify registration
  • Check that the target agent_id is correct (format: computer_id:session_id)
  • Use check_messages to manually poll

Example Multi-Agent Workflow

On Mac (Session A):

You: Ask all agents to report their hostname

Claude: [uses broadcast tool]
Broadcast sent to 2 agents: "Please report your hostname"

On Mac (Session B) - auto-injected:

╔════════════════════════════════════════════════════════════════╗
║  📬 INCOMING MESSAGES FROM OTHER AGENTS                        ║
╠════════════════════════════════════════════════════════════════╣
║  [5] From: Csabas-Mac-Pro.local:6da26f26 → session:abc12345
║  Message: Please report your hostname
╚════════════════════════════════════════════════════════════════╝

You: (any prompt)

Claude: I see a message asking for my hostname. Let me reply.
[uses reply tool with content: "Hostname: Csabas-Mac-Pro.local"]

On Linux Server - auto-injected:

╔════════════════════════════════════════════════════════════════╗
║  📬 INCOMING MESSAGES FROM OTHER AGENTS                        ║
║  [6] From: Csabas-Mac-Pro.local:6da26f26 → session:def67890
║  Message: Please report your hostname
╚════════════════════════════════════════════════════════════════╝

You: handle the message

Claude: [uses reply tool with content: "Hostname: linux-server"]

Back on Mac (Session A):

You: check messages

Claude: You have 2 replies:
- [7] From Csabas-Mac-Pro.local:abc12345: "Hostname: Csabas-Mac-Pro.local"
- [8] From linux-server:def67890: "Hostname: linux-server"

Docker

Run the Agent Hub server in a Docker container with one command. The database persists across restarts.

Quick Start

# Build and start in one step
docker run -d --name agent-hub -p 8765:8765 -v hub-data:/app/data csabakecskemeti/agent-hub:latest

Using docker-compose (Recommended)

# Start the server
docker-compose up -d

# View logs
docker-compose logs -f agent-hub

# Stop the server
docker-compose down

# Restart with fresh database
docker-compose down -v && docker-compose up -d

Custom Port

Override the host port with the HUB_PORT environment variable:

HUB_PORT=9000 docker-compose up -d
# Access at http://localhost:9000/agents

Build & Push (for maintainers)

docker build -t agent-hub:latest .
docker tag agent-hub:latest csabakecskemeti/agent-hub:latest
docker push csabakecskemeti/agent-hub:latest

License

MIT

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选