Connect your cameras. Give your agent eyes.

MCP overview

Rikuy is an MCP server. Any agent that speaks MCP can add it with one URL and ask the owner to approve the connection.

server     https://mcp.rikuy.si/mcp
transport  Streamable HTTP, stateless
spec       MCP 2026-07-28
auth       OAuth 2.1 + PKCE, clients by CIMD
card       https://mcp.rikuy.si/mcp/server-card
registry   si.rikuy/cameras

How a look works

1. Your agent calls list_cameras and picks a camera by its "what it sees".
2. It calls get_frame, get_frames or get_clip with a time and a purpose.
3. Rikuy checks, on the server: membership active, location not cut, camera on, hour allowed, days back. If anything fails, it returns a named error.
4. Rikuy records the look before it asks for any image.
5. The Bridge brings the frames from the recorder, fixes their time and uploads them. Every answer carries the note "Review the clip before acting on it."
6. The agent answers the owner, then calls record_answer with its conclusion, and flags it if it found something.

What the server asks of agents

• Describe facts. Do not identify people by their face.
• Stop and tell the owner if an image shows a restroom or changing area.
• Call record_answer after answering, so the owner sees the conclusion next to the look.

Cost to the owner's AI plan

Each image uses tokens from the owner's AI plan. Rikuy sends frames at 1280 × 720 and caps a single look at 24 frames. As a reference, Claude counts about 1,200 tokens for one such frame.

Coming in v1.1

Resources rikuy://locations and rikuy://locations/{loc}/cameras, and a prompt review_night_shift for convenience stores.