unifi-protect — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited unifi-protect (Agent Skill) and scored it 96/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 1 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Requires: protect Optional: network
Verify required tools through MCP discovery before collecting data; discovery alone does not prove controller connectivity or authorization. If a required server/tool is missing or a required read fails, stop and report it unavailable. Optional sources may be skipped only when the report visibly names the missing source and resulting coverage limits. Use the product setup skill for connection help; never request secrets.
Every report must include a Coverage and Limitations section: name each requested source and its status (available, unavailable, partial, or capped), requested time window/timezone or current-state collection time, actual filters, returned counts, limits, pagination, and any failed calls or unknown fields. Counts describe retrieved records, not totals unless completeness is established. A successful empty query is an observation; a failed or missing source is unavailable, never an empty result. If a limit is reached, label the source capped unless further bounded reads establish coverage. If completeness cannot be established, label it partial. Do not claim complete coverage, all-clear, or overall health from unavailable, partial, or capped data; say “No concerning activity found in the retrieved records” when appropriate.
Separate Observations (cite source/tool, record ID and timestamp when available), Hypotheses (correlations and alternative explanations), and Identity (only explicit source-backed credential/account attribution, with its limits). Timing, similar names, proximity, or absence of a matching event alone proves neither identity nor cause. A credential event identifies the recorded account, not who physically used it. Missing mappings or clock uncertainty limit correlations.
You have access to a UniFi Protect MCP server that lets you query and manage a UniFi Protect NVR. It provides 63 tools covering cameras, smart detections, Find Anything detection search, recordings, snapshots, lights, sensors, chimes, Known Faces, license plates, and the Alarm Manager (arm/disarm).
The server uses lazy loading by default — only meta-tools are registered initially:
| Meta-Tool | Purpose |
|---|---|
protect_tool_index | Discover tools by name/description; use category, search, or include_schemas to filter |
protect_execute | Call any tool by name (essential in lazy mode) |
protect_batch | Run multiple tools in parallel |
protect_batch_status | Check async batch job status |
Workflow: Call protect_tool_index to find the right tool, then protect_execute to call it. Use protect_batch for multiple independent queries.
All mutations are disabled by default because Protect controls physical security hardware.
Read operations — always available. Listing cameras, events, snapshots, sensor readings — all work without permissions.
Mutations require explicit opt-in via env vars:
UNIFI_POLICY_PROTECT_CAMERAS_UPDATE=true — camera settings, recording toggle, PTZ, rebootUNIFI_POLICY_PROTECT_LIGHTS_UPDATE=true — light brightness, PIR sensitivityUNIFI_POLICY_PROTECT_CHIMES_UPDATE=true — chime volume, triggerUNIFI_POLICY_PROTECT_ALARM_UPDATE=true — arm/disarm the Alarm Manager (Protect 6.1+)UNIFI_POLICY_PROTECT_RECOGNITION_UPDATE=true — Known Face rename/mergeUNIFI_POLICY_PROTECT_RECOGNITION_DELETE=true — Known Face deletionThe category segment is the server's config key (CAMERAS, LIGHTS, CHIMES, SENSORS), not the singular permission_category shorthand in tools_manifest.json (camera, light, chime). UNIFI_POLICY_PROTECT_CAMERA_UPDATE is never read. A denied tool's error names the exact variable to set.
Confirmation flow — every mutation uses preview-then-confirm:
confirm=true → executes the mutationAlways preview first and show the user before confirming.
All tools return: {"success": true, "data": ...}, {"success": false, "error": "..."}, or {"success": true, "requires_confirmation": true, "preview": ...}. Always check success first.
Redacted secrets: RTSP/RTSPS stream aliases and URLs come back as ***REDACTED*** by default from protect_get_camera_streams. Raw values are controlled by process policy (UNIFI_PROTECT_REDACT_SENSITIVE_FIELDS=false or global UNIFI_REDACT_SENSITIVE_FIELDS=false), not by tool arguments.
protect_get_snapshot with include_image=true returns base64 JPEG inlineprotect_get_camera_streams gives stream URL metadata for video player integration; raw URLs require redaction policy to be disabled for a trusted local processprotect_list_smart_detections filters by type (person, vehicle, animal, package, face, licensePlate). These are the highest-signal events — prioritize over raw motion.protect_detection_search_labels to discover controller-supported label values, then pass those values to protect_search_detections for richer searches by vehicle type, color, device, or other Protect labels.camera; use protect_list_cameras to resolve missing names rather than assuming every event contains camera_name.protect_recent_events reads from websocket buffer instantly (no API call). Buffer holds ~100 events with 5-minute TTL. Use protect_list_events for historical queries.protect_export_clip returns metadata (not video data — too large for MCP). Max 2 hours, supports timelapse (fps: 4=60x, 8=120x, 20=300x)protect_ptz_preset with saved positionsprotect_list_known_faces to inspect face groups before rename, merge, or delete mutationsvalue strings.protect_list_events with time range filters.limit parameter to keep responses focused.security-digest skill which handles batch calls, severity classification, and cross-product correlation.Username and password are required (local admin credentials, not Ubiquiti SSO). A UniFi Protect API key is optional and enables selected capabilities implemented through the Protect Integration API, including sensor settings, per-camera chime ring settings, and viewer liveview assignment.
To configure, run /unifi-protect:unifi-protect-setup or set env vars manually:
UNIFI_PROTECT_HOST=192.168.1.1
UNIFI_PROTECT_USERNAME=admin
UNIFI_PROTECT_PASSWORD=your-passwordIf the user also has networking or door access control, other UniFi MCP plugins are available:
unifi-network — network devices, clients, firewall, VPN, routingunifi-access — door locks, credentials, visitors, access policiesCameras are network clients — if a camera appears offline, the Network server can help check connectivity via unifi_lookup_by_ip.
For the complete list of all 63 tools organized by category with descriptions, tips, and common scenarios, read references/protect-tools.md.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.