facetrak-mcp
Skill0 starsChanged 5 months agoMITView on GitHub
Use this skill when the user mentions face tracking, face recognition, face scanning, servo tracking, face detection with camera, or anything related to the FaceTrak project. This skill tells an LLM how to use the FaceTrak MCP server (stdio-based) to query and control a real-time face detection, recognition, and pan-tilt tracking system. FaceScan exposes 20 tools via MCP (Model Context Protocol) over stdio. The server wraps: - facedb — HOG-based face recognition with .npy encoding storage - engine — MediaPipe…
In this repository2 other files it gives its agents.
- opencvSkill
- python-opencvSkill
The file
Files
FaceTrak MCP — LLM Interface for Face Tracking
Use this skill when the user mentions face tracking, face recognition, face scanning, servo tracking, face detection with camera, or anything related to the FaceTrak project. This skill tells an LLM how to use the FaceTrak MCP server (stdio-based) to query and control a real-time face detection, recognition, and pan-tilt tracking system.
Architecture
FaceScan exposes 20 tools via MCP (Model Context Protocol) over stdio. The server wraps:
facedb— HOG-based face recognition with.npyencoding storageengine— MediaPipe face detection + OpenCV camera loopservo— Serial pan/tilt controller (Arduino)pose— Head pose estimation via FaceLandmarks + solvePnPrecorder— Video recording to MP4config— JSON config persistence
Start the server:
just mcp-serveor: facetrak-mcp CLI command or python -m facetrak.mcp_server
In Claude Desktop config:
{
"mcpServers": {
"facetrak": {
"command": "/path/to/.venv/bin/python",
"args": ["-m", "facetrak.mcp_server"]
}
}
}Tools Overview
Face Database (no camera needed)
| Tool | When to use |
list_faces | User asks "who is registered", "show me known faces", "list people" |
get_face_info(name) | User asks about a specific person's encoding/details |
forget_person(name) | User says "delete X", "remove Y", "forget Z" |
register_person(name) | User says "add new person", "register X", "learn this face" — starts camera, captures samples, registers |
Camera Control
| Tool | When to use |
camera_list | User asks "which cameras", "what sources", "find cameras", or to discover available USB/IP cameras |
camera_add(name, source) | User wants to add an IP camera (RTSP URL) or name a USB camera |
camera_switch(index) | User wants to switch to a different camera by index from camera_list |
start_camera | User wants to start tracking or see the camera feed |
stop_camera | User wants to stop tracking or release the camera |
get_status | User asks "what's happening", "is it running?", current pan/tilt/pose, active camera |
toggle_recording | User wants to start/stop recording video |
toggle_blur | User wants to enable/disable privacy blur on unknown faces |
Configuration
| Tool | When to use |
get_config | User asks "show settings", "what's configured" |
update_config(key, value) | User wants to change a setting (e.g. detection width, recognition threshold, servo limits) |
reset_config | User wants factory defaults |
Servo Control
| Tool | When to use |
servo_list_ports | User asks "what ports are available", "find my Arduino" |
servo_connect(port) | User wants to connect to a servo controller |
servo_disconnect | User wants to disconnect |
servo_set_enabled(enabled) | User wants to turn tracking on/off |
servo_set_angle(pan, tilt) | User wants to set a specific angle (0-180) |
Important Notes
- Face DB is persistent — stored as
.npyfiles infaces/data/. Tools likelist_faces,forget_person,get_face_infowork immediately without camera. - Camera is on-demand —
start_cameraopens the camera and starts background polling.register_persontemporarily starts/stops camera if needed. - Multi-camera —
camera_listdetects all USB cameras + shows configured ones.camera_addadds RTSP URLs or IP cameras.camera_switchchanges live source. - Registration —
register_person(name)opens camera, watches for 2 seconds collecting face crops, then saves. If no face is visible, it fails. - Config keys — flat keys like
detect_width,recog_threshold,camera,blur_unknown, or dotted servo keys likeservo.port,servo.pan_min,servo.baud. - Servo angles — clamped to configured min/max (default 0-180). Send
P090T045\nstyle protocol over serial. - Head pose — yaw/pitch/roll available via
get_statuswhen camera is running.
Typical Workflows
"Who do you know?" → list_faces → print the list
"Register Alice" → register_person("Alice") → wait 2s → confirm
"Start tracking" → start_camera → servo_list_ports → servo_connect("/dev/cu.usbmodem...") → servo_set_enabled(true) → get_status
"Switch to IP camera" → camera_add("Garage", "rtsp://192.168.1.100:554/stream") → camera_switch(1) → get_status
"What cameras are available?" → camera_list → shows all USB + configured cameras
"What's the current status?" → get_status → shows pan/tilt, pose, recording state, face position
"Delete Bob" → forget_person("Bob") → confirm
"Change detection resolution to 640" → update_config("detect_width", "640")
"Record video" → start_camera → toggle_recording → ...later... → toggle_recording
"Turn on privacy blur" → start_camera → toggle_blur → on
"Move servo to center" → servo_set_angle(90, 90)
Skill Metadata
- Skill name: facetrak-mcp
- Trigger on: face tracking, face recognition, face detection, face scan, servo tracking, camera tracking, person registration, "who is this", "recognize face"
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

