Integrations & MCP
CCTV / VSaaS exposes its data programmatically through MCP (Model Context Protocol) — the same integration surface used across Performance Hub. The preferred way to build on the platform is through the AI Agent appliance, which speaks MCP natively, holds facility context, and can react to events in real time.
Two ways in
| Path | What it is | When |
|---|---|---|
| The AI Agent appliance (preferred) | Your facility's AI Agent connects to Performance Hub over MCP out of the box. Ask it questions ("who was in the facility at 6 this morning?"), build automations that react to events, and let it correlate CCTV data with everything else it knows | Almost everything — no code, real-time, and conversational |
| Direct MCP | Any MCP-capable client or agent connecting with a Performance Hub MCP key | Custom software, external agent platforms, scripted integrations |
The AI Agent path is preferred because events flow to it as they happen — a person detection, a device going offline — so integrations built there are reactive, not polled. The details of connecting an AI Agent are in Connecting to Performance Hub.
Authentication and scoping
Direct MCP access authenticates with a personal MCP key (a phk_… token) generated from your Performance Hub profile. Every tool call is scoped by the same rules as the UI:
- You can only query facilities your account can access.
- Tools are gated by module: the vision and people tools require the facility to have CCTV and the people features enabled; device tools require Device Management.
- Calls are subject to the same audit posture as the rest of the platform.
Treat MCP keys like passwords — scope them to service accounts where possible and rotate them when staff change.
The tool surface
The full, always-current catalogue of MCP tools — names, parameters, and what each returns — lives in the AI Agents MCP documentation, so there is a single source of truth to consult. For this module, the relevant categories are:
- CCTV tools — the vision and people-data surface: searching the detection timeline, looking up people and their visit activity, check-in history, and facility report snapshots. (The data these return belongs to AI Insights & People; the tools are catalogued there because they're the programmatic surface of the camera platform.)
- Device Management tools — listing Edge Processors and cameras with their status, and sending remote commands to devices.
The same MCP surface carries the rest of Performance Hub — door and gate control, user profile and facility lookup, and more — so an integration can span modules: "when an unknown person is detected after hours, check the door logs and notify the manager" is one AI Agent automation, not three systems.
Patterns that work well
- After-hours monitoring — an AI Agent automation on detection events outside opening hours, notifying staff with the camera and timestamp.
- Daily digest — a scheduled automation summarising yesterday: visits, busiest hours, camera health, upload backlog.
- Fleet monitoring at scale — multi-facility operators polling the device-listing tools across facilities from their NOC tooling, or letting an AI Agent watch and escalate.
- Investigation assistant — "show me everyone who entered between 5 and 6 pm" answered conversationally through the detection-timeline search tools, instead of manual scrubbing.
What integrations can't do
The MCP surface is data and control-plane access. Video itself — live streams and recording playback — is served to browsers through the module and guest share links, not through MCP. For handing footage to an external party, generate a share link.
Related pages
- AI Agent — Agent capabilities and Automations
- Security & privacy — the audit posture integrations inherit
- Access control & permissions