Pyats Network Automation

🚀 pyATS MCP Server

This project implements a Model-based Contextual Protocol (MCP) server that encapsulates Cisco pyATS and Genie functionality. It enables structured, model-based network device interactions using the JSON-RPC 2.0 protocol over STDIO.

🚨 This server does not use HTTP or SSE. All communications occur via STDIN/STDOUT, making it ideal for secure embedded, containerized, or LangGraph tool integrations.

✨ Features

• Connect to Cisco IOS/NX-OS devices (defined in the pyATS testbed file).
• Support secure execution of verified CLI commands (show, ping).
• Allow controlled configuration changes.
• Return structured (parsed) or raw output.
• Expose a set of predefined tools via tools/discover and tools/call.
• Run entirely over STDIO for maximum portability with minimal surface area.

🚀 Quick Start

Prerequisites

Make sure Python and Docker are installed.

Clone the Repository

git clone https://github.com/your-repository.git

Run the Server

Set the Testbed Path

export PYATS_TESTBED_PATH=/absolute/path/to/testbed.yaml

Continuous STDIO Mode (Default)

python3 pyats_mcp_server.py

This starts a long-running process that reads JSON-RPC requests from standard input and writes responses to standard output.

One-shot Mode

echo '{"jsonrpc": "2.0", "id": 1, "method": "tools/discover"}' | python3 pyats_mcp_server.py --oneshot

This processes a single JSON-RPC request and then exits.

Docker Support

Build the Container

docker build -t pyats-mcp-server .

Run the Container (STDIO Mode)

docker run -it --rm -v $(pwd):/app -e PYATS_TESTBED_PATH=/app/testbed.yaml pyats-mcp-server

⚠️ Important Note

The testbed.yaml file is compatible with Cisco Modeling Labs (CML) Sandbox on Cisco DevNet!

💻 Usage Examples

MCP Tools

The following tools are available:

LangGraph Integration

You can configure it in LangGraph as follows:

from pyats.topology import loader

# Load the testbed file
testbed = loader.load("testbed.yaml")

# Create an MCP server instance
mcp_server = testbed.devices["iosxe"].connections.mcp

# Add it to the LangGraph pipeline
graph.add_node(mcp_server, label="MCP Server")

🔧 Technical Details

Security Features

• All inputs are verified and sanitized.
• Execution of unsafe commands (e.g., reload or shutdown) is blocked.
• Support for SSL/TLS encrypted communication is coming soon.

Project Structure

.
├── pyats_mcp_server.py   # Main program of the MCP server
├── testbed.yaml         # Testbed configuration file containing Cisco device connection information
└── Dockerfile          # Docker image build file

MCP Server Configuration

The MCP server can be configured in the following JSON format:

{
  "devices": {
    "iosxe": {
      "hostname": "192.168.1.1",
      "username": "admin",
      "password": "password"
    }
  }
}

Complete LangGraph Graph Configuration Example

from pyats.topology import loader
from langgraph.core.graph import Graph

# Load the testbed file
testbed = loader.load("testbed.yaml")

# Create an MCP server instance
mcp_server = testbed.devices["iosxe"].connections.mcp

# Create a LangGraph manager
graph = Graph()

# Add nodes and edges
graph.add_node(mcp_server, label="MCP Server")

📄 License

No license information provided in the original document.

✍️ Author

John Capobianco

Product Marketing Evangelist at Selector AI

Author of Automate Your Network

If you have any questions, feel free to contact me. I'm happy to help with:

• Adding a LangGraph graph configuration example.
• Providing a配套客户端脚本 (Note: The original Chinese term is kept as no clear English equivalent is given).
• CI/CD integration (e.g., GitHub Actions).