LW MCP Agents

LW MCP Agents is a lightweight, modular framework for building and orchestrating AI agents using the Model Context Protocol (MCP). It empowers you to rapidly design multi-agent systems where each agent can specialize, collaborate, delegate, and reasonwithout writing complex orchestration logic.

Build scalable, composable AI systems using only configuration files.

---

Why Use LW MCP Agents?

• Plug-and-Play Agents: Launch intelligent agents with zero boilerplate using simple JSON configs.
• Multi-Agent Orchestration: Chain agents together to solve complex tasksno extra code required.
• Share & Reuse: Distribute and run agent configurations across environments effortlessly.
• MCP-Native: Seamlessly integrates with any MCP-compatible platform, including Claude Desktop.

---

What Can You Build?

• Research agents that summarize documents or search the web
• Orchestrators that delegate tasks to domain-specific agents
• Systems that scale reasoning recursively and aggregate capabilities dynamically

---

Architecture at a Glance

---

Table of Contents

• Getting Started
• Example Agents
• Running Agents
• Custom Agent Creation
• How It Works
• Technical Architecture
• Acknowledgements

---

Getting Started

Installation

git clone https://github.com/Autumn-AIs/LW-MCP-agents.git
cd LW-MCP-agents
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

Run Your First Agent

python src/agent/agent_runner.py --config examples/base_agent/base_agent_config.json

Try a Multi-Agent Setup

Terminal 1 (Research Agent Server):

python src/agent/agent_runner.py --config examples/orchestrator_researcher/research_agent_config.json --server-mode

Terminal 2 (Orchestrator Agent):

python src/agent/agent_runner.py --config examples/orchestrator_researcher/master_orchestrator_config.json

Your orchestrator now intelligently delegates research tasks to the research agent.

---

Claude Desktop Integration

Configure agents to run inside Claude Desktop:

1. Locate your Claude config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

2. Add your agent under mcpServers:

{
  "mcpServers": {
    "research-agent": {
      "command": "/bin/bash",
      "args": ["-c", "/path/to/venv/bin/python /path/to/agent_runner.py --config=/path/to/agent_config.json --server-mode"],
      "env": {
        "PYTHONPATH": "/path/to/project",
        "PATH": "/path/to/venv/bin:/usr/local/bin:/usr/bin"
      }
    }
  }
}

---

Example Agents

• Base Agent
  A minimal agent that connects to tools via MCP.
  examples/base_agent/

• Orchestrator + Researcher
  Demonstrates hierarchical delegation and capability sharing.
  examples/orchestrator_researcher/

Contribute your own example! Submit a PR or reach out to the maintainers.

---

Running Agents

Basic Command

python src/agent/agent_runner.py --config <your_config.json>

Advanced Options

Option	Description
--server-mode	Exposes the agent as an MCP server
--server-name	Assigns a custom MCP server name

---

Custom Agent Creation

Minimal Config

{
  "agent_name": "my-agent",
  "llm_provider": "groq",
  "llm_api_key": "YOUR_API_KEY",
  "server_mode": false
}

Adding Capabilities

Define specialized functions the agent can reason over:

"capabilities": [
  {
    "name": "summarize_document",
    "description": "Summarize a document in a concise way",
    "input_schema": {
      "type": "object",
      "properties": {
        "document_text": { "type": "string" },
        "max_length": { "type": "integer", "default": 200 }
      },
      "required": ["document_text"]
    },
    "prompt_template": "Summarize the following document in {max_length} words:\n\n{document_text}"
  }
]

Orchestrator Agent

{
  "agent_name": "master-orchestrator",
  "servers": {
    "research-agent": {
      "command": "python",
      "args": ["src/agent/agent_runner.py", "--config=research_agent_config.json", "--server-mode"]
    }
  }
}

---

How It Works

Capabilities as Reasoning Units

Each capability:

1. Fills in a prompt using provided arguments
2. Executes internal reasoning using LLMs
3. Uses tools or external agents
4. Returns the result

Research Example

[INFO] agent:master-orchestrator - Executing tool: research_topic
[INFO] agent:research-agent - Using tool: brave_web_search
[INFO] agent:research-agent - Finished capability: research_topic

---

Technical Architecture

Key Components

Component	Role
AgentServer	Starts, configures, and runs an agent
MCPServerWrapper	Wraps the agent to expose it over MCP
CapabilityRegistry	Loads reasoning tasks from config
ToolRegistry	Discovers tools from other agents

Architecture Highlights

• Hierarchical Design: Compose systems of agents with recursive reasoning
• Delegated Capabilities: Agents delegate intelligently to peers
• Tool Sharing: Tools available in one agent become accessible to others
• Code-Free Composition: Create entire systems via configuration

---

Acknowledgements

This project draws inspiration from the brilliant work on mcp-agents by LastMile AI.