UniFi Network - Claude Code Skill

Unofficial. Community-built Claude Code Skill and MCP server for the UniFi Network integration API. Not affiliated with, endorsed by, or sponsored by Ubiquiti Inc.

Prerequisites: Install the CLI

This skill drives the unifi-network-cli binary. You must verify the CLI is installed before invoking any command from this skill. If it is missing, install it first:

1. macOS / Linux:

   bash <(curl -fsSL https://raw.githubusercontent.com/Servosity/msp-skills/main/skills/unifi-network/install.sh)
   

2. Windows (PowerShell):

   iwr -useb https://raw.githubusercontent.com/Servosity/msp-skills/main/skills/unifi-network/install.ps1 | iex
   

3. Verify: unifi-network-cli --version
4. Ensure ~/.local/bin (macOS / Linux) or %LOCALAPPDATA%\Programs\msp-skills (Windows) is on $PATH.

If --version reports "command not found" after install, the install step did not put the binary on $PATH. Do not proceed with skill commands until verification succeeds.

unifi-network-cli wraps the full local Network integration API (devices, clients, firewall, ACL, networks, VPN, switching) with a local SQLite mirror. That mirror is what lets it answer questions the live API can't: what changed since yesterday, what device just joined, and which firewall rule would match a given packet.

When to Use This CLI

Use this CLI for scripted or agent-driven UniFi Network audits, firewall/ACL review, and change-detection on a self-hosted gateway. Built for MSP owners and technicians who want terminal-first, agent-drivable control over a client's UniFi controller instead of clicking the web UI.

Anti-triggers

Do not use this CLI for:

• Do not use this for UniFi Protect (cameras) or UniFi Access (doors) - this CLI only covers the Network integration API.
• Do not use this for the legacy cloud/CloudKey Controller API - that's a different auth model entirely.
• Do not treat rule-predict's output as a live-gateway guarantee - it simulates against the last synced ruleset.

Unique Capabilities

These capabilities aren't available in any other tool for this API.

Local state that compounds

• topology - Group every synced client under the device it is attached to, a device-to-client tree built entirely from local mirror data, no live crawl needed.

  Reach for this when an agent needs to see which clients sit behind which device without walking every device endpoint individually.

  unifi-network-cli topology --site default --json
  

• drift - Show what changed in site config (networks, firewall, wifi, DNS) since the last drift run for this site.

  Use after a suspected config change to see exactly what moved, without manually diffing the controller UI.

  unifi-network-cli drift --site default --since 24h --json
  

  --since is echoed in the output for reference only - the comparison is always against the snapshot the previous drift run captured, not a fixed lookback window. The first run for a site captures the baseline and reports no changes.

• newcomer - List devices and clients first seen since a given sync, for spotting new hardware joining the network.

  Use for periodic security review of what joined the network recently.

  unifi-network-cli newcomer --since 7d --json
  

Agent-native plumbing

• port-audit - Review port utilization and PoE status across every switch on a site. Pass --json for per-port detail; the plain terminal path prints one N up / M down, PoE active on K port(s) line per device. It reads the device list from the local mirror but fetches interfaces live, one call per switching or gateway device.

  Use before adding new PoE devices to check headroom, or to find unused ports across a stack.

  unifi-network-cli port-audit --site default --json
  

• guest report - Summarize guest network usage: active vouchers and connected guest clients, from local data.

  Use for a quick guest-network health check without cross-referencing separate voucher and client screens. Note the output contains live voucher codes.

  unifi-network-cli guest report --site default --json
  

• rule-predict - Predict which firewall policy would match a hypothetical packet before making a live change. Matches on source and destination IP only: --port is echoed for reference and is not used for matching. Pass host IPs, not CIDRs - a CIDR's mask is ignored and only the address you typed is tested, so --src 10.0.3.0/24 predicts for the single address 10.0.3.0 and tells you nothing about 10.0.3.50. It does not evaluate the range.

  Use to check the effect of a proposed firewall change before applying it live.

  unifi-network-cli rule-predict --src 10.0.3.50 --dst 10.0.0.1 --port 443 --json
  

Paginated reads default to 25 rows

Nearly every paginated read takes --limit (default 25, except sites hotspot get-vouchers which defaults to 100) and --all. This is not limited to the sites ... subtree: countries, pending-devices, dpi get-application-categories, and dpi get-applications behave the same way. Offset pagination emits no truncation warning, so page 1 comes back looking like the whole answer. Any question phrased as "every" or "all" needs --all:

unifi-network-cli sites devices get-adopted-overview-page <siteId> --all
unifi-network-cli sites firewall get-policies <siteId> --all

Before any local-mirror command: sync

drift, newcomer, topology, guest report, rule-predict, and search all read the local SQLite mirror, and port-audit needs the synced device list before it can fetch port detail. On a fresh install the mirror is empty:

unifi-network-cli sync

An unsynced mirror looks like an empty network, not like an error. Only the explicit --site <name> form fails loudly (exit 3, site "..." not found in local mirror). Without --site - the natural call on a single-site gateway, since the flag defaults to "the only synced site" - every one of these exits 0 with an empty payload and puts the run 'unifi-network-cli sync' first hint on stderr only:

Command (no --site, empty mirror)	Exit	stdout
newcomer --json	0	[]
port-audit --json	0	[]
topology --json	0	{"site":"","devices":[],"unattached_clients":[]}
guest report --json	0	empty vouchers + guests
drift --json	0	{"site":"","first_run":false,"changes":[]}
rule-predict --json	0	"matched":false

drift is the sharpest edge: it positively asserts first_run:false, changes:[] - "not a first run, nothing changed" - for a gateway that was never synced. Always run sync first, and read stderr; never report "nothing found" from stdout alone.

Command Reference

countries - Manage countries

• unifi-network-cli countries - Returns ISO-standard country codes and names, used for region-based configuration or regulatory compliance.

dpi - Manage dpi

• unifi-network-cli dpi get-application-categories - Returns predefined Deep Packet Inspection (DPI) application categories used for traffic identification and filtering.
• unifi-network-cli dpi get-applications - Lists DPI-recognized applications grouped under categories. Useful for firewall or traffic analytics integration.

info - Manage info

• unifi-network-cli info - Retrieve general information about the UniFi Network application.

pending-devices - Manage pending devices

• unifi-network-cli pending-devices - Retrieve a paginated list of devices pending adoption, including basic device information.

sites - Endpoints for listing and managing UniFi sites within a local Network application. Site ID is required for most other API requests.

• unifi-network-cli sites - Retrieve a paginated list of local sites managed by this Network application.

This reference is partial - it lists the top-level convenience commands only. The full surface is 137 commands, and almost all of it lives under sites ... (devices, clients, firewall, ACL, networks, DNS, WiFi, hotspot, switching, VPN, WANs, RADIUS, traffic-matching-lists), including every write and destructive command.

Enumerate it with --help, recursively:

unifi-network-cli sites --help                 # the 14 resource groups
unifi-network-cli sites firewall --help        # that group's commands

Do not rely on api for discovery: it only lists interfaces tagged as API resources, which in this CLI is dpi alone (2 of 137 commands). guide.md is likewise a partial reference, not the full surface.

Finding the right command

When you know what you want to do but not which command does it, ask the CLI directly:

unifi-network-cli which "<capability in your own words>"

which resolves a natural-language capability query against this CLI's curated feature index. That index holds only the 6 novel commands (topology, drift, newcomer, port-audit, guest report, rule-predict) - not the 131 others. A question about an API command can therefore score a spurious hit and exit 0: which "list every adopted device" returns topology, not sites devices get-adopted-overview-page. Use which to find the novel commands, and --help for everything else. Exit code 0 means at least one match, 2 means no confident match; under --agent or --json the exit code is always 0, so test matches for emptiness instead.

Recipes

Find who just joined the network

unifi-network-cli newcomer --since 7d --json --select id,name,mac

Narrow a potentially large newcomer list down to just the fields needed to identify each device.

Audit switch port headroom before adding a PoE device

unifi-network-cli port-audit --site default --json

Lists PoE status and free ports across every switch on the site.

Check what a firewall change would match

unifi-network-cli rule-predict --src 10.0.3.50 --dst 10.0.0.1 --port 443 --json

Simulates rule evaluation order against the synced ruleset before making a live change.

Auth Setup

Two environment variables are required:

• UNIFI_API_KEY - generate a local API key from the gateway's own UI (Settings -> Control Plane -> Integrations -> Create API Key).
• UNIFI_GATEWAY_HOST - the gateway's hostname or IP on your network, e.g. 10.0.0.1. The CLI builds the base URL from it (https://<host>/proxy/network); the integration spec declares no absolute server, so there is no default to fall back on. Set UNIFI_BASE_URL instead to point at a non-standard endpoint directly.

The gateway's self-signed certificate is handled automatically for private, loopback, and link-local hosts, so the common case needs no extra configuration. There is no --insecure flag; the auto-detection matches only a literal RFC1918 / loopback / link-local IP or the exact string localhost - any DNS name (unifi.lan included) is treated as not-private, so a hostname or public IP will fail certificate verification and needs UNIFI_INSECURE_SKIP_VERIFY=1 set deliberately. It also accepts 0/false to force verification back on when you have installed a real certificate.

Run unifi-network-cli doctor to check config, paths, and gateway reachability. It does not validate the credential - it reports present, not verified and exits 0 even for a wrong key (a 401 still shows API: reachable). To confirm the key actually works, run a read command such as unifi-network-cli sites and check for exit 4.

Agent Mode

Add --agent to any command. Expands to: --json --compact --no-input --no-color --yes.

• Pipeable - JSON on stdout, errors on stderr

• Filterable - --select keeps a subset of fields. Dotted paths descend into nested structures; arrays traverse element-wise. Critical for keeping context small on verbose APIs:

  unifi-network-cli countries --agent --select code,name
  

• Previewable - --dry-run shows the request without sending

• Offline-friendly - sync/search commands can use the local SQLite store when available

• Non-interactive - never prompts. Note that most sites ... commands take required POSITIONAL arguments rather than flags (sites firewall get-policies <siteId>, sites devices execute-port-action <siteId> <deviceId> <portIdx>); there is no --site-id flag

• Explicit retries - use --idempotent only when an already-existing create should count as success, and use --ignore-missing only when a missing delete target should count as success

Response envelope

Under --agent, commands that read from the local store or the API wrap output in a provenance envelope. Plain --json does not wrap - it emits the bare payload for backward compatibility, so .results and .meta are absent. The examples in this file use --json and therefore return bare JSON; add --agent if you want the envelope:

{
  "meta": {"source": "live" | "local", "synced_at": "...", "reason": "..."},
  "results": <data>
}

Parse .results for data and .meta.source to know whether it's live or local. A human-readable N results (live) summary is printed to stderr only when stdout is a terminal AND no machine-format flag (--json, --csv, --compact, --quiet, --plain, --select) is set - piped/agent consumers and explicit-format runs get pure JSON on stdout.

Paths and state

Agents should treat the CLI's path resolver as part of the runtime contract:

• Use --home <dir> for one invocation, or set UNIFI_HOME=<dir> to relocate all four path kinds under one root.

• Use per-kind env vars only when a specific kind must diverge: UNIFI_CONFIG_DIR, UNIFI_DATA_DIR, UNIFI_STATE_DIR, UNIFI_CACHE_DIR.

• Resolution order is per-kind env var, --home, UNIFI_HOME, XDG (XDG_CONFIG_HOME, XDG_DATA_HOME, XDG_STATE_HOME, XDG_CACHE_HOME), then platform defaults.

• config contains settings like config.toml and profiles. data contains credentials.toml, data.db, and novel-snapshots/ (drift's per-site snapshots and newcomer's first-seen file). state contains teach.log, learnings.jsonl, and the learn journal. The .agentcookie-managed marker sits beside config.toml in the CONFIG dir, not data. cache contains regenerable HTTP/cache files.

• Stored secrets live in credentials.toml under the data dir. Existing legacy config.toml secrets are read for compatibility and leave config.toml on the first auth write.

• Run unifi-network-cli doctor --fail-on warn to surface path and credential-location warnings. agent-context exposes a schema v4 paths block for agents that need the resolved dirs.

• For MCP, pass relocation through the MCP host config. The MCP binary does not inherit CLI flags:

  {
    "mcpServers": {
      "unifi": {
        "command": "unifi-network-mcp",
        "env": {
          "UNIFI_HOME": "/srv/unifi"
        }
      }
    }
  }
  

Fleet precedence: an inherited per-kind env var overrides an explicit --home for that kind. Use UNIFI_HOME or per-kind vars as durable fleet levers, and use --home only for a single invocation. Relocation is not reversible by unsetting env vars; move files manually before clearing UNIFI_HOME, or doctor will not find credentials left under the former root.

Automatic learning

This CLI ships a self-capturing learning loop. The CLI does its own bookkeeping: every invocation is journaled locally, a failed flag followed by a corrected retry auto-derives a flag_alias candidate, and a teach on a query family without a playbook auto-synthesizes a playbook_candidate from the session's journal. Your job is judgment only: recall first, act on surfaced candidates, teach the final answer, playbook amend when you observe a correction. You never record failures by hand.

Step 1: recall before any discovery

Before list/search/drill commands on a new user question, run:

unifi-network-cli recall "<user's question>" --agent

The response envelope. NOTE: every field below is nested under a top-level results object (alongside meta), so the hit array is .results.results and the flag is .results.found - not .found:

{
  "query": "...",
  "normalized": "<normalized form>",
  "query_entities": ["..."],
  "found": true | false,
  "match_score": 0.0,
  "results": [
    { "resource_id": "...", "resource_type": "...", "venue": "...",
      "confidence": 2, "entity_match": "exact|partial|unknown",
      "source": "taught|preseed|pattern", "warnings": ["..."] }
  ],
  "mismatches": [ /* only when --debug-mismatches */ ],
  "warnings": [ /* top-level */ ],
  "candidates": [
    { "id": 12, "class": "flag_alias | playbook_candidate",
      "summary": "...", "sightings": 3, "last_seen": "...",
      "rationale": "...",
      "next_action": ["<trial command>", "unifi-network-cli learnings confirm 12"] }
  ],
  "playbook": {
    "query_family": "...",
    "playbook": {
      "steps": [ { "cmd": "<command with {slot} substitution>", "purpose": "..." } ],
      "entity_slots": ["$ENTITY"],
      "expected_tool_calls": 3
    },
    "slots_resolved": { "$ENTITY": { "token": "<live token>", "canonical": "<canonical>" } },
    "notes": "<workarounds + gotchas for this query family>"
  },
  "notes": "<duplicate surface for non-playbook callers>"
}

Empty-store short-circuit: if the store has no learnings, playbooks, or candidates yet (recall finds nothing and learnings list and learnings candidates are both empty), skip recall for the rest of this session instead of taxing every query; resume recall-first once something has been taught.

Step 2: decision tree

Read candidates, playbook, notes, results[0], and warnings in that order:

if Candidates present (warnings include "candidates_present"):
    -> candidates are try-then-confirm, never facts. Follow each candidate's
       two-step next_action verbatim: run the trial command first, then run
       `learnings confirm <id>` only after the trial verified the behavior.
       Reject a wrong candidate with `learnings reject <id>`.
    -> NEVER re-teach something recall surfaced as a candidate; confirm or
       reject that candidate instead of teaching a duplicate.
    -> candidates ride alongside playbooks and resource hits, not instead of
       them; continue with the branches below after acting on them.

if Playbook present:
    -> READ Playbook.notes verbatim FIRST (workarounds + gotchas the CLI surface doesn't expose)
    -> replay Playbook.steps in order, substituting Playbook.slots_resolved entries
       for the entity slot tokens. If a step's slot is unresolved, fall back to
       discovery for that step only.
    -> the Playbook's expected_tool_calls is a budget; if you find yourself running
       materially more, record the divergence via `unifi-network-cli playbook amend`
       at end-of-session.

elif Notes present (no Playbook):
    -> read Notes verbatim before any discovery step; they carry known gotchas
       for this query family even when no structured choreography exists yet.

elif Found AND Results[0].EntityMatch == "exact" AND Results[0].Confidence >= 2:
    -> skip discovery; fetch live data for Results[*].ResourceID in parallel

elif Found AND Results[0].EntityMatch == "partial":
    -> candidate hint, NOT a hit; read the resource title to validate before trusting

elif (any row in Mismatches[] when --debug-mismatches was passed):
    -> treat as cold start; the stored learning is for a different entity
       (different canonical resolved from query_entities)

else:  // Found == false, no playbook, no notes
    -> cold start; run discovery normally; teach the answer afterward (Step 4).
       If the family has no playbook yet, that teach auto-synthesizes a
       playbook candidate from this session's journal - you do not need to
       record one by hand.

Playbook and Notes are orthogonal to the per-resource path. A recall response can carry both a Playbook AND a Results[] hit - use both: the Playbook tells you which choreography to run; the resource hits short-circuit specific steps. Default to skipping mismatches; pass --debug-mismatches only when investigating cold-start surprises.

Candidate judgment details: learnings confirm <id> prints the candidate's full payload before materializing it - check that the printed payload matches the behavior you verified. learnings reject <id> tombstones the derivation signature so the same candidate does not resurface. The envelope carries only the few candidates worth acting on now; unifi-network-cli learnings candidates lists the full open set.

Graceful degradation: if learnings confirm is an unknown command, you are driving an older binary - ignore the candidates guidance and follow the rest of the protocol.

Step 3: always read warnings

• low_confidence: row exists at confidence<2. Treat as a hint, not a skip-discovery hit.
• resource_not_in_store: the local store doesn't have the resource the learning points at. The match validator couldn't classify entities - direct-fetch and re-evaluate.
• cross_alias_match (per-result): the row was taught under a different alias and matched the live query's canonical via entity_lookups (e.g., a "USA" teach satisfying a "United States" recall). Trust the resource_id.
• similar_shape_different_entity:<canonical> (top-level): a structurally matching row exists but its canonical entity differs from the live query's. Treated as cold start; the warning carries the conflicting canonical as a hint, but the row is NOT promoted into Results.
• ambiguous_alias (top-level): a single query entity resolved to multiple canonicals (e.g., a device name that exists on two sites). Surface the ambiguity from context before committing to a resource.
• candidates_present (top-level): the envelope carries a candidates section. Handle it via the candidates branch in Step 2 before anything else.
• lookup_refresh_available (top-level): an entity in the query has no lookup row yet, but synced data could provide one. Run unifi-network-cli sync to refresh entity lookups.
• Top-level no_learnings_for_query_family: the table had no rows above the Jaccard floor. Pure cold start.

Step 4: teach & after finalizing your response - always

Teaching is unconditional. After resolving a query the store could not answer, background-teach the final resource mapping - no call-count threshold, no judging whether it was "worth" learning. The teach is the anchor of the loop: it triggers playbook synthesis for a family without a playbook, and same-referent phrasings fold into one family so near-duplicate teaches do not fragment the store. Fire it after assembling your user-facing response but BEFORE emitting it, with a shell & so the call returns immediately:

unifi-network-cli teach --query "<user's question>" --resource-type <type> --resource <id1> --resource <id2>
# (append shell `&` to background it)

Silent on success. Errors only land in teach.log under the resolved state dir. Teach the most specific resource - if the user asked a broad question and you walked through parent records to find the specific answer, teach the leaf id, not the parent. Cross-alias resolution uses the entity_lookups table. This CLI ships no seed data for it (the generated initLearn is a no-op), so it starts empty and fills only from sync and from explicit teach-lookup rows - a teach under one alias does not resolve another alias until a lookup row exists. Use teach-lookup when a device or site is known by more than one name.