drunk-ai-client 0.1.0

STDIO MCP bridge client for Drunk AI Proxy

Drunk MCP Client (STDIO)

A stdio MCP client that connects to the Drunk MCP Proxy server, enabling remote MCP services with agent synchronization and remote resource management.

Features

• Remote Agent Sync: Automatically sync agents from remote sources
• Skill Management: Local skill directory synchronization
• Resource Downloader: Fetch and cache remote MCP resources
• Agent Execution: Execute agents with permission controls
• Offline Support: Cached agents for offline use

Installation & Usage

Local Development (from repo root)

uvx --from ./src/drunk_ai_client drunk_ai_client

Published Package

uvx drunk_ai_client

Direct Installation

pip install -e ./src/drunk_ai_client
drunk-ai-client

Configuration

The client supports multiple configuration sources:

Environment Variables

Variable	Required	Description
DRUNK_API_URL	Yes	Drunk MCP Proxy server URL (e.g., http://localhost:9123)
DRUNK_API_KEY	No	Bearer token for authentication
DRUNK_SKILL_DIR	No	Local directory for synced skills (default: ./skills)
DRUNK_AGENTS_DIR	No	Local directory for synced agents (default: ./agents)
DRUNK_SYNC_INTERVAL	No	Sync interval in seconds (default: 300)
DRUNK_ALLOW_OVERWRITE	No	Overwrite local files during sync (true/false)

Command Line Options

drunk-ai-client --sync --agent-source https://remote.repo/agents.yaml

New Agent Sync Feature

Sync Remote Agents

# Sync from a remote source (new!)
DRUNK_API_URL=http://localhost:9123 \
DRUNK_AGENTS_DIR=./my-agents \
drunk-ai-client --sync-agents --source https://your-agent-repo.com/agents.yaml

Agent Directory Structure

agents/
├── agent1/
│   ├── config.yaml
│   ├── skills/
│   └── resources/
└── agent2/
    └── ...

Skills Sync

# Sync skills from proxy
DRUNK_SKILL_DIR=./skills drunk-ai-client --sync-skills

Usage Examples

Connect to Proxy with Agent Sync

# Start the client with agent sync enabled
DRUNK_API_URL=http://localhost:9123 \
DRUNK_AGENTS_DIR=./agents \
drunk-ai-client --sync --agent-source https://repo.com/agents.yaml

Execute a Remote Agent

Once agents are synced locally, they can be used as standard MCP tools:

// Claude Code or any MCP client can use these tools
{
  "name": "agent_search",
  "description": "Search through synced agents",
  "inputSchema": { ... }
}

Development

Run Tests

python -m pytest tests/

Build and Install Locally

pip install -e ".[dev]"

Notes

• The stdio bridge logs to stderr by default (FastMCP behavior)
• Agent sync runs in background if --sync flag is enabled
• Remote resources are cached locally with automatic refresh
• Authentication tokens are validated on each request
• Agents execute with sandboxed permissions

See Also

• Drunk MCP Proxy Server
• Agent Provider Documentation
• Configuration Guide