openskycam
An open-source sky camera system that continuously records night-sky footage, automatically detects and classifies moving targets such as meteors using AI, and supports AI-driven operation through the MCP protocol.
Timeline (last 90 days)
The developer built an open-source sky camera system called openskycam and let Claude Code take over the entire process from detection to classification.
On Saturday night, the openskycam system captured 17 meteors.
The openskycam system records 4K sky footage 24/7, and all footage is saved on a local disk.
The openskycam workflow is responsible for extracting frames, running the detection model, linking detections into tracks to distinguish moving from stationary targets, and sorting hits into three buckets: meteors, aircraft and fixed light sources.
Every feature on the openskycam website is simultaneously an API call, and a built-in MCP server lets Claude Code drive the entire system directly.
Claude built a nightly workflow that runs automatically whenever a new recording arrives.
Claude reviews each frame's detection results, accepting or rejecting them one by one.
When the model kept misclassifying stars as flying objects, Claude built a dataset from the detection results the author had approved and retrained the model.
On night data it had never been trained on, the star false-positive rate dropped from about half to 7%, while still capturing 90% of meteors.
The system discards invalid results every morning and only pushes valuable meteors to the phone; on overcast nights it stays quiet and sends no messages.
7 more timeline events
All Facts (17)
Some version of the system has been observing the author's sky since 2024, and the UFO count remains zero.
50%UnverifiedThe developer built an open-source sky camera system called openskycam and let Claude Code take over the entire process from detection to classification.
50%UnverifiedThe openskycam system records 4K sky footage 24/7, and all footage is saved on a local disk.
50%UnverifiedThe openskycam workflow is responsible for extracting frames, running the detection model, linking detections into tracks to distinguish moving from stationary targets, and sorting hits into three buckets: meteors, aircraft and fixed light sources.
50%UnverifiedEvery feature on the openskycam website is simultaneously an API call, and a built-in MCP server lets Claude Code drive the entire system directly.
50%UnverifiedClaude built a nightly workflow that runs automatically whenever a new recording arrives.
50%UnverifiedClaude reviews each frame's detection results, accepting or rejecting them one by one.
50%UnverifiedWhen the model kept misclassifying stars as flying objects, Claude built a dataset from the detection results the author had approved and retrained the model.
50%UnverifiedOn night data it had never been trained on, the star false-positive rate dropped from about half to 7%, while still capturing 90% of meteors.
50%UnverifiedThe system discards invalid results every morning and only pushes valuable meteors to the phone; on overcast nights it stays quiet and sends no messages.
50%Unverifiedopenskycam's MCP server ships with guides and skills (build-a-workflow, find-detections, verify-changes), and Claude learns the rules from the server.
50%Unverifiedopenskycam's guardrails are enforced by the server: no key can create new keys, delete operations first show a preview and require confirmation, and edit operations fail outright rather than overwrite when a newer version is encountered.
50%Unverifiedopenskycam can be installed unattended via `npm run setup -- --yes`, reading all configuration from .env, and it stops and identifies the missing item when a setting is absent; however, the Cloudflare tunnel, login app and API keys still need to be created manually.
50%UnverifiedRunning openskycam requires an RTSP camera, a Raspberry Pi, a machine with a large-capacity disk (macOS or Linux, GPU optional), and a Cloudflare account with a domain.
50%UnverifiedThe openskycam project involves two repositories: openskycam handles the camera, detection and the tools Claude uses, and edmund-harness handles the iMessage messaging channel.
50%UnverifiedThe system's daily video runs from 9:30 p.m. to 6 p.m. the next day; at night every meteor and aircraft comes with actual footage from the 4K camera, while during the day each identified bird is marked with a pink dot, with a real bird close-up attached every 10 detections.
50%UnverifiedOn Saturday night, the openskycam system captured 17 meteors.
50%