network-aiops
Disclaimer: This is a community-maintained open-source project and is not affiliated with, endorsed by, or sponsored by Cisco, Arista, Juniper, NetBox Labs, or any network vendor. Vendor and product names are trademarks of their respective owners. Source code is publicly auditable at github.com/AIops-tools/Network-AIops under the MIT license.
Governed multi-vendor network device operations for AI agents — 33 MCP tools,
every one wrapped with the bundled @governed_tool harness: a local unified audit
log under ~/.network-aiops/, token/runaway budget guard, undo-token recording,
and descriptive risk-tier labels. Credentials (device passwords + the NetBox token)
are kept in an encrypted store (secrets.enc), never plaintext on disk.
Devices are reached over NAPALM; an optional NetBox block adds source-of-truth lookups.
Standalone: the governance harness is bundled in the package (
network_aiops.governance) — network-aiops has no external skill-family dependency. Coverage focuses on common device operations and is not yet exhaustive.
Verification status: the test suite is mock-based; not yet validated against live devices — self-testable with cEOS / vMX / containerlab. See docs/VERIFICATION.md.
What works
Read device facts, interfaces (+ counters/IP), BGP/LLDP neighbors (summary and
detail), ARP/MAC tables, VLANs, route lookups, hardware environment, optics, NTP,
users, SNMP info, VRFs, and an aggregated device_health; run read-only RCA
diagnostics that flag down/erroring/flapping interfaces and unhealthy BGP
neighbors — each finding citing the measured number that tripped it; back up the
running config, dry-run a config diff, and merge/replace/rollback config — across the five
core NAPALM platforms below. Optional NetBox lookups (devices + interfaces) confirm
intended state before a change.
NAPALM does not implement every getter on every platform; an unsupported getter
returns a teaching error ("not supported by the <driver> driver") rather than
crashing. Secrets are never returned — get_users redacts password hashes and
get_snmp_information redacts community strings.
Supported devices
Additional platforms (Nokia SR OS / SR Linux, Huawei VRP, etc.) are reachable via NAPALM community drivers but are not officially tested here. Need one? See Contributing.
Supported actions
What this tool does, and does not, decide
It delivers multi-vendor network device (NAPALM) + NetBox operations — reads and writes — accurately and efficiently, and records every one of them. It does not decide whether a write is allowed to happen. That is the agent's judgement, or the permission of the account you connect it with: log in with a device account at a read-only privilege level (and give NetBox a read-only API token), and the writes fail at the server — the place that actually owns the permission.
So there is no read-only switch, no policy file, no approval gate to configure. The one thing the
tool guarantees is that nothing is silent: every call, over MCP and over the CLI alike, lands an
audit row in ~/.network-aiops/audit.db, and destructive writes still capture their before-state
and record an inverse where one exists.
Each tool declares a
risk_level, carried into the audit row as a descriptive tier (none/confirm/review) — so a reviewer can see at a glance that a row was a high-risk delete. It is a label, not a gate.
Running a smaller / local model? See agent-guardrails.md — it lists the guardrails this tool enforces for you (so you don't spend prompt budget restating them) and gives a ready-made system prompt for what's left.
Quick Start
As a Claude Code plugin
One install gives an agent both the skill and the MCP server:
/plugin marketplace add AIops-tools/marketplace
/plugin install network-aiops@aiops-tools
The MCP server is fetched with uv and pinned to the
package version this plugin declares, so an audit row can be traced back to the
code that wrote it. Credentials are still configured with network-aiops init — see below.
As an OpenClaw plugin
The same bundle is published on ClawHub, where one install delivers the skill and its MCP server together:
openclaw plugins install clawhub:@zw008/network-aiops
openclaw skills info network-aiops # expect: Visible to model: yes
Restart the OpenClaw gateway afterwards so it loads the plugin. The MCP server is
fetched with uv, pinned to this exact release, so
uvx has to be on PATH — without it the skill still installs but reports
Visible to model: no. Credentials are configured exactly as below.
As a CLI or standalone MCP server
uv tool install network-aiops
network-aiops init # wizard: device + driver + host + encrypted password
network-aiops doctor
network-aiops device facts -t core-sw1
network-aiops device health -t core-sw1
network-aiops diagnose interface-health -t core-sw1 # worst-first interface RCA
network-aiops diagnose bgp -t core-sw1 # worst-first BGP-neighbor RCA
network-aiops config backup -t core-sw1 -o core-sw1.cfg
Playbook: triage a flaky uplink before touching config
# 1. Ask the device what's actually wrong — findings are ranked worst-first and
# each cites the measured value (error count, last-flap seconds, uptime).
network-aiops diagnose interface-health -t core-sw1
network-aiops diagnose bgp -t core-sw1
# 2. If interface-health flags a link admin-up/oper-down with climbing errors and
# BGP shows the peer on that path recently reset, you have your root cause: a
# physical-layer fault (cable/optic) resetting the session — not routing.
# 3. Confirm intended state, then remediate with the governed, audited path.
network-aiops device counters -t core-sw1
network-aiops config diff -t core-sw1 -f fix.cfg # dry-run the change first
Create ~/.network-aiops/config.yaml:
devices:
- name: core-sw1 # used as -t core-sw1
driver: eos # ios | nxos | nxos_ssh | iosxr | eos | junos
host: 10.0.0.1
username: admin
optional_args: # passed verbatim to NAPALM (optional)
secret: enable-pw # enable/secret
port: 443
# Optional source-of-truth:
netbox:
url: https://netbox.example.com
Secrets are stored encrypted in ~/.network-aiops/secrets.enc (Fernet/AES +
scrypt-derived key; chmod 600) — never in config.yaml or a plaintext .env.
Device passwords are keyed by device name; the NetBox token uses the reserved
name netbox-token:
network-aiops init # interactive wizard (recommended)
network-aiops secret set core-sw1 # store a device password (hidden prompt)
network-aiops secret set netbox-token # store the NetBox API token
network-aiops secret list # names only — values are never printed
network-aiops secret migrate # import a legacy plaintext .env, then delete it
Export NETWORK_AIOPS_MASTER_PASSWORD to unlock the store non-interactively (MCP
server / cron). Legacy plaintext env vars (NETWORK_<TARGET_UPPER>_PASSWORD,
NETWORK_NETBOX_TOKEN) remain a deprecated fallback. An empty device password is
allowed for key-based SSH auth.
MCP
{
"command": "network-aiops",
"args": ["mcp"],
"env": {
"NETWORK_AIOPS_CONFIG": "~/.network-aiops/config.yaml",
"NETWORK_AIOPS_MASTER_PASSWORD": "…" // unlocks the encrypted secret store
}
}
Audit & Safety
- Every tool call is logged to
~/.network-aiops/audit.db(local SQLite; relocate withNETWORK_AIOPS_HOME). config_merge/config_replacecapture the pre-change running config and record an inverseconfig_replace-to-backup undo descriptor.config_replaceisrisk_level=high; CLI destructive commands (config merge/replace/rollback) require double confirmation and support--dry-run(which prints the diff without committing).- All device text passes through
sanitize()(output hygiene: control/format-char stripping + truncation). - Device passwords and the NetBox token live only in the encrypted
secrets.enc(chmod 600); tools never return passwords, SNMP community strings, or hashes.
See skills/network-aiops/SKILL.md and SECURITY.md for details.
Companion Skills
Contributing & feature requests
Coverage is intentionally focused. Need a device or action that isn't here yet? Open an issue or pull request at github.com/AIops-tools/Network-AIops — contributions, feature requests, and comments are all welcome.