Agent Workflows

This project implements agentic workflows using LangChain, LangGraph, and the Model Context Protocol (MCP).

Prerequisites

• Python 3.13+
• Docker & Docker Compose (optional, for containerized execution)

Installation

1. Clone the repository:

   git clone <repository-url>
   cd agent-workflows

2. Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies:

   pip install .

4. Configure environment: Copy .env.example to .env and fill in the required values.

   cp .env.example .env

Running the Application

Local

python src/agents/math/math_agent.py

NOT IMPLEMENTED WIP:

python src/main.py

Docker

docker compose up --build

Adding a New MCP Client

To add a new Model Context Protocol (MCP) client to the application, you need to register it in the configuration.

1. Open src/config/mcp_config.py.
2. Locate the AppConfig.load method.
3. Add a new entry to the services dictionary using McpServiceConfig.

Example:

If you want to add a "weather" service running on port 8080:

# src/config/mcp_config.py

# ... inside AppConfig.load method ...

# Weather Service
services["weather"] = McpServiceConfig(
    server_url=os.getenv("MCP_WEATHER_URL", "http://127.0.0.1:8080/weather/weather"),
    redirect_uris=["http://127.0.0.1:8080/weather/"]
)

4. (Optional) Add the corresponding environment variable MCP_WEATHER_URL to your .env file to allow configuration overrides.