Agent integration
MCP video tools preview for World-State, evidence, timeline, and cost.
This page explains the intended tool surface for agent frameworks. It is not a production MCP activation page. The working beta path remains REST, dashboard, and artifacts.
MCP video tools preview
| Tool | Input | Output | Boundary |
|---|---|---|---|
| get_scene_graph | video_id | objects, events, zones, confidence, evidence_refs | read-only |
| get_timeline | video_id | ordered event list and semantic refresh markers | read-only |
| get_evidence | video_id, evidence_ref | time span, description, review notes | read-only |
| ask_video | video_id, question, require_evidence=true | answer, citations, uncertainty, review flag | no unsupported claims |
| get_cost_context | video_id | duration, detector mode, fallback, retention, estimated cost | budget visibility |
Framework wrapper sketch
def mcp_tool_get_scene_graph(video_id: str, user):
authorize_read_only(user.tenant_id, video_id)
artifact = ayneye.get_scene_graph(video_id)
return {{
"tool": "get_scene_graph",
"read_only": True,
"artifact_version": artifact.version,
"data": artifact.data,
"blocked_actions": ["identity", "intent", "physical_action"]
}}
def mcp_tool_ask_video(video_id: str, question: str, user):
answer = ayneye.ask_video(video_id, question, require_evidence=True)
if not answer.evidence_refs:
return {{"answer": "insufficient_evidence", "review_required": True}}
return answerWhat is intentionally not enabled
No public beta MCP tool unlocks doors, controls cameras, dispatches emergency response, identifies people, infers intent, deletes evidence, or sends automated outreach. Those operations require separate governance, customer policy, audit, and human review paths.
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.