worker-agent 1.0.4

A config-driven, plug-and-play worker agent template built on LangGraph + FastMCP.

🤖 Universal Worker Agent Template

A config-driven, plug-and-play worker agent template.
Clone the folder, edit config.yaml, and you have a brand new specialized agent — no code changes needed.

---

Architecture

┌─────────────────────────────────────────────┐
│               Main Agent                    │
│   (calls workers via MCP tool calls)        │
└───────────────────┬─────────────────────────┘
                    │ MCP (stdio / SSE)
          ┌─────────▼──────────┐
          │      main.py       │  ◄─── FastMCP bridge
          │  execute_task(...)  │       exposes 1 tool
          └─────────┬──────────┘
                    │ asyncio
          ┌─────────▼──────────┐
          │   core/agent.py    │  ◄─── LangGraph ReAct loop
          │  LangGraph + Ollama │
          └─────────┬──────────┘
                    │ MCP clients (stdio)
        ┌───────────┼───────────┐
   ┌────▼────┐  ┌───▼────┐  ┌──▼──────┐
   │FS Server│  │  Web   │  │ Custom  │
   │  (npx)  │  │ Search │  │ Server  │
   └─────────┘  └────────┘  └─────────┘

Component	File	Responsibility
Config	config.yaml	Defines everything — identity, model, tools, server port
Config loader	core/config_loader.py	Parses YAML into typed dataclasses
ReAct Agent	core/agent.py	Loads MCP tools, runs LangGraph loop
MCP Bridge	main.py	Exposes execute_task() as an MCP server

---

Quick Start

1. Install uv (if not already installed)

# Windows (PowerShell)
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

2. Install dependencies

uv sync

This creates a .venv and installs all dependencies from pyproject.toml automatically.

For development extras (ruff, mypy, pytest):

uv sync --all-extras

3. Configure your agent

Edit config.yaml — the only file you need to touch:

agent:
  name: "FileMaster"
  description: "A specialized worker that organizes files and manages directory structures."
  system_prompt: "You are an expert at organizing and refactoring local files."

model:
  provider: "ollama" # Supported: "ollama", "openai", "gemini"
  model_name: "llama3.2" # any model loaded in Ollama
  # api_key: "your-api-key-here" # Uncomment if using openai or gemini (or set API_KEY env var)

mcp_clients:
  - name: "filesystem-server"
    command: "npx"
    args:
      ["-y", "@modelcontextprotocol/server-filesystem", "C:/Users/Dev/Project"]

server:
  transport: "stdio" # or "sse"
  port: 8001

4. Run as a subprocess (stdio) — standard MCP

uv run worker-agent

5. Run as an HTTP server (SSE) — call it from a browser or remote agent

uv run worker-agent --transport sse --port 8001

---

Cloning a New Worker

1. Copy the whole folder:

   cp -r Agent_a Agent_researcher
   

2. Edit only config.yaml in the copy:
   • Change agent.name, agent.description, agent.system_prompt
   • Swap out mcp_clients for the tools this worker needs
   • Change server.port so it doesn't conflict
3. Install & run:

   cd Agent_researcher
   uv sync
   uv run worker-agent
   

---

Example Worker Configs

File Specialist

agent:
  name: "FileMaster"
  description: "A specialized worker that organizes files and manages directory structures."
  system_prompt: "You are an expert at organizing and refactoring local files."
mcp_clients:
  - name: "filesystem-server"
    command: "npx"
    args:
      ["-y", "@modelcontextprotocol/server-filesystem", "C:/Users/Dev/Project"]

Web Researcher

agent:
  name: "SearchPro"
  description: "A research agent to summarize technical documentation found on the web."
  system_prompt: "You specialize in deep web research and summarizing technical docs."
model:
  provider: "openai"
  model_name: "gpt-4o-mini"
  api_key: "sk-..."
mcp_clients:
  - name: "brave-search"
    command: "npx"
    args: ["-y", "@modelcontextprotocol/server-brave-search"]

Local Python MCP Server

mcp_clients:
  - name: "my-server"
    command: "python"
    args: ["D:/DEV/mcp/server.py"]

---

How the Main Agent Calls This Worker

In your Main Agent's MCP config:

{
  "mcpServers": {
    "file-worker": {
      "command": "uv",
      "args": ["--directory", "D:/DEV/mcp/Agent_a", "run", "worker-agent"]
    }
  }
}

Dynamic Configuration (Multiple Workers)

If your Main Agent needs multiple specialized workers, you do not need to clone the entire repository multiple times. You can create several configuration YAML files (e.g., web_worker.yaml, file_worker.yaml) and pass them dynamically using the WORKER_AGENT_CONFIG environment variable.

{
  "mcpServers": {
    "web-worker": {
      "command": "uv",
      "args": ["--directory", "D:/DEV/mcp/Agent_a", "run", "worker-agent"],
      "env": {
        "WORKER_AGENT_CONFIG": "D:/DEV/mcp/Agent_a/web_worker.yaml"
      }
    },
    "file-worker": {
      "command": "uv",
      "args": ["--directory", "D:/DEV/mcp/Agent_a", "run", "worker-agent"],
      "env": {
        "WORKER_AGENT_CONFIG": "D:/DEV/mcp/Agent_a/file_worker.yaml"
      }
    }
  }
}

Or run it directly with uvx and an environment variable if the package is published:

# Linux/macOS
WORKER_AGENT_CONFIG="./custom_config.yaml" uvx worker-agent

# Windows (PowerShell)
$env:WORKER_AGENT_CONFIG="./custom_config.yaml"; uvx worker-agent

The worker exposes one tool:

• execute_task(instruction: str) → str

The Main Agent calls it like any other MCP tool. The worker handles reasoning, tool use, and error recovery internally, returning only the final result.

---

Project Structure

Agent_a/
├── config.yaml           ← The only file you edit per clone
├── main.py               ← MCP bridge (FastMCP)
├── pyproject.toml         ← All deps & build config (hatchling)
├── README.md
└── core/
    ├── __init__.py
    ├── agent.py          ← LangGraph ReAct loop
    └── config_loader.py  ← YAML → typed dataclasses

---

Development Commands

# Install in dev mode with all extras
uv sync --all-extras

# Run from the project directory
uv run worker-agent
uv run worker-agent --transport sse --port 8001

# Build a distributable wheel + sdist
uv build

# Publish to PyPI
uv publish