aiNOC

aiNOC is a Python-based MCP server that implements an On-Call watcher workflow for a multi-vendor network troubleshooting framework. It monitors Vector's log output for SLA/path failure events, automatically opens a Jira ticket, and invokes Claude-powered agents to diagnose and propose fixes. The system orchestrates agent sessions through the oncall/watcher.py script, which coordinates with the MCP tools and the defined skills to guide operators toward validation and application of fixes. Users interact with the watcher in interactive mode during development or rely on the service mode for production deployments where the watcher runs persistently, spawning agent sessions in a tmux-managed environment for each incident.

Prerequisites:

• Python 3.8+ installed on the host
• Virtual environment support (venv) available
• Access to the aiNOC repository code

Step-by-step installation:

1. Clone the repository

git clone https://github.com/pdudotdev/aiNOC/
cd aiNOC

2. Create and activate a Python virtual environment

python3 -m venv mcp
source mcp/bin/activate

3. Upgrade pip and install dependencies

pip install --upgrade pip
pip install -r requirements.txt

4. Configure environment variables (see notes in additional_notes) and ensure Vector and Jira integrations are accessible as per your environment.
5. Run the watcher in the appropriate mode (interactive for development, service for production):

# Interactive mode (development)
python3 oncall/watcher.py

# Production service mode would typically be: create and enable a systemd service for oncall-watcher

Tips and common considerations:

• Ensure Python dependencies in requirements.txt are satisfied and compatible with your environment.
• Vector must be configured to emit logs at /var/log/network.json so the watcher can detect incidents.
• Jira integration requires credentials/configuration; verify API access and token validity before deployment.
• If running as a service, consider using tmux to manage individual agent sessions as described in aiNOC documentation.
• Review and customize CLAUDE.md and /skills directories to tailor troubleshooting workflows for your network.
• Maintain security hygiene for environment variables and credentials; avoid committing secrets to version control.
• If the watcher fails to start, check permissions on /var/log/network.json and ensure the user running the process has read access.
• For troubleshooting, run in interactive mode first to observe agent session behavior before enabling production service mode.