Scheduled network change and compliance digests with a headless AI agent
A morning digest answers the questions every network lead asks: what changed overnight, and is anything out of policy? A headless agent on a schedule can read rConfig over MCP, write a short summary, and post it before anyone logs in.
Best for: team leads, daily stand-ups, weekly management reporting.
Tools: rConfig, a headless agent, a scheduler (cron, CI schedule or n8n).
rConfig tools used: get_devices, get_config_history, get_config_diff, get_compliance_status.
Choosing a runner
| Runner | How it runs | Guide |
|---|---|---|
| Claude Code | claude -p "..." --allowedTools "mcp__rconfig__*" |
Claude Code |
| Codex CLI | codex exec "..." with default_tools_approval_mode = "writes" |
Codex CLI |
| n8n | Schedule trigger, AI Agent with the rConfig MCP Client Tool | n8n |
A daily change digest
#!/usr/bin/env bashset -euo pipefail# RCONFIG_MCP_TOKEN comes from the scheduler's secret store.
claude -p "Using rConfig, for every device in the Core, Edge and WAN groups:find running-config changes in the last 24 hours (compare last_config_at first,then check history only for devices that changed). For each change, one line:device, time, what changed. Then give the compliance summary for each group.Under 250 words. No preamble." \ --allowedTools "mcp__rconfig__*" > digest.txt
# Post digest.txt to your chat webhook, email it, or attach it to a ticket.The prompt tells the agent to check last_config_at on each device first. Only devices that changed need a history lookup, which keeps the run fast and cheap.
An illustrative digest:
Overnight changes (24h)- dub-core-01, 02:14: VLAN 30 removed from Gi1/0/48 trunk- lon-edge-02, 23:40: two ACL 120 entries added for 10.40.0.0/16
ComplianceCore 96% (1 failing: dub-core-03, SSH v1 enabled)Edge 100%WAN 88% (2 not evaluated: no backup in 3 days)A weekly management digest
Change the scope and the tone: a week of changes counted per group, the compliance trend, and the three items that need a decision. Ask for plain language with no config syntax.
Limits to know
- Scale. Config history is read per device. Scope digests by group or tag rather than “all devices”.
- Stale backups look like no change. Devices that haven’t backed up recently show no changes. Ask the agent to list devices whose
last_config_atis older than the digest period. - Keep it read-only. The job needs rConfig’s tools and somewhere to post. Don’t give the scheduled agent any other servers.
Related reference
- Connect an AI agent over MCP
- Policy compliance
- Scheduled tasks: for the backups the digest depends on
Frequently asked questions
How do I schedule an AI agent to report on network changes?
Run a headless agent such as Claude Code (claude -p) or Codex (codex exec) from cron, a CI schedule or n8n, with rConfig's MCP server connected and its read-only tools pre-approved. Send the output to chat, email or a ticket.
How many devices can a digest cover?
History is read one device at a time, so a digest over hundreds of devices makes hundreds of calls. Scope it by group or tag, or split it into several digests.
What token should a scheduled job use?
A token for a dedicated rConfig user whose role covers only the devices in the digest, stored in your scheduler's secret store.