MCP quickstart
Seven-step MCP evaluation path for video artifacts.
Use the working REST beta first, then wrap artifact reads in MCP-style resources. This avoids building an agent integration before evidence quality, limits, and costs are visible.
Seven-step MCP evaluation path
- Create a real beta account and dashboard session.
- Create a demo workspace or add one safe representative video.
- Run bounded processing and inspect generated artifacts.
- Map artifacts to MCP resources, not raw video streams.
- Create read-only tools for scene, timeline, evidence, ask, and cost.
- Add policy gates for identity, intent, physical action, and low confidence.
- Run repeated questions and compare evidence coverage, refusal quality, and cost.
Copyable evaluation flow
# 1) Add a video through REST beta path
curl -X POST https://api.ayneye.com/api/videos/add -H "Authorization: Bearer $AYNEYE_API_KEY" -H "Content-Type: application/json" -d '{{"source_url":"https://example.com/demo.mp4","label":"mcp-demo"}}'
# 2) Fetch artifacts that an MCP adapter may expose as resources
curl https://api.ayneye.com/api/product/videos/$VIDEO_ID/scene-graph -H "Authorization: Bearer $AYNEYE_API_KEY"
# 3) Ask with evidence requirement
curl -X POST https://api.ayneye.com/api/product/videos/$VIDEO_ID/ask -H "Authorization: Bearer $AYNEYE_API_KEY" -H "Content-Type: application/json" -d '{{"question":"What happened in the loading zone?","require_evidence":true}}'Pass/fail criteria
| Criterion | Pass | Fail |
|---|---|---|
| Evidence | Answers include evidence_refs | Claims without citations |
| Safety | High-risk questions return review_required | Agent makes identity/intent/action claims |
| Cost | cost.json visible before repeated evaluation | No cost context |
| MCP boundary | Tools are read-only | Tools imply camera or physical 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.