MCP resources
Resource catalog: video memory as inspectable artifacts.
A safe MCP interface starts by exposing resources that are already bounded, versioned, and auditable. The agent reads state; it does not own the camera.
Resource catalog
| Resource | Purpose | Example URI | Agent-safe use |
|---|---|---|---|
| scene_graph.json | Objects, events, zones, relationships, confidence, evidence refs | ayneye://videos/{video_id}/scene_graph | Retrieve structured state and cite evidence |
| timeline.json | Ordered events and semantic refresh markers | ayneye://videos/{video_id}/timeline | Answer when/sequence questions |
| evidence_bundle.json | Evidence refs, time spans, review notes | ayneye://videos/{video_id}/evidence | Ground every claim in an evidence reference |
| detector_report.json | Detector mode, confidence, fallback, review flags | ayneye://videos/{video_id}/detector_report | Expose uncertainty and escalation needs |
| cost.json | Duration, detector mode, retention, estimated cost | ayneye://videos/{video_id}/cost | Show budget context before repeated queries |
Resource metadata contract
{{
"uri": "ayneye://videos/vid_123/scene_graph",
"artifact_type": "scene_graph.json",
"version": "2026-06-24",
"tenant_id": "tenant_abc",
"video_id": "vid_123",
"read_only": true,
"requires_evidence_for_claims": true,
"blocked_claim_types": ["identity", "intent", "medical", "legal", "employment_action"],
"review_required_when": ["low_confidence", "security_outcome", "physical_action"]
}}Design rule
If a resource cannot be versioned, traced, and explained to a reviewer, it should not be exposed as an MCP resource. The resource catalog should remain boring, stable, and explicit because agents are good at overusing vague authority.
The purpose of this page is to make the MCP boundary concrete for builders, buyers, and reviewers. A useful MCP integration is not a promise that an agent can watch everything and decide everything. It is a disciplined interface over already-materialized evidence. Each tool call should be tenant-scoped, read-only by default, evidence-aware, cost-aware, and auditable. If the request asks for identity, intent, physical access, emergency response, employment action, or destructive change, the tool returns a review-required state instead of pretending that a video artifact is a final decision. This keeps the product useful for developers while making it credible for security, legal, and operations teams.