MindBotz CorridorKey RunPod Workstation

MindBotz-branded RunPod/docker workstation fork of edenaion/EZ-CorridorKey, the desktop/noVNC GUI wrapper for CorridorKey.

This fork keeps the upstream GUI and Docker/noVNC workflow, then adds a small workstation control layer for remote/cloud use:

• Browser GUI via noVNC on port 6080
• File upload/browser UI on port 6081
• MindBotz HTTP API on port 8080
• MindBotz MCP-style JSON-RPC endpoint on port 8765
• stdio MCP module for local agent integration
• RunPod-focused compose/template notes
• persistent volumes for clips, outputs, models, projects, config, and logs

Attribution and License

This repo is derived from edenaion/EZ-CorridorKey and keeps upstream attribution. Upstream license is CC BY-NC-SA 4.0. Treat this fork as non-commercial unless/until licensing is reviewed and permission is obtained for any commercial deployment.

Quick Start: Docker Compose GPU

cd docker
MINDBOTZ_WORKSTATION_PASSWORD='change-me' \
MINDBOTZ_API_TOKEN='optional-api-token' \
docker compose -f docker-compose.runpod.yml up -d --build

Open:

Service	URL	Notes
GUI/noVNC	http://localhost:6080	password = MINDBOTZ_WORKSTATION_PASSWORD
Upload UI	http://localhost:6081	user admin, password = MINDBOTZ_WORKSTATION_PASSWORD
API health	http://localhost:8080/health	bearer auth only if MINDBOTZ_API_TOKEN is set
MCP HTTP	http://localhost:8765	JSON-RPC endpoint for initialize, tools/list, tools/call
Raw VNC	localhost:5900	optional; avoid exposing publicly

RunPod Template Setup

Best practical path:

1. Build and push this image to a registry you control:

   docker build -f docker/Dockerfile -t your-dockerhub/mindbotz-corridorkey-runpod:latest .
   docker push your-dockerhub/mindbotz-corridorkey-runpod:latest

2. Create a RunPod template using that image.

3. Expose HTTP ports:

   • 6080 for noVNC GUI
   • 6081 for file browser
   • 8080 for API
   • 8765 for MCP HTTP JSON-RPC

4. Optional TCP port:

   • 5900 raw VNC, only if you know you need it

5. Recommended env vars:

   MINDBOTZ_WORKSTATION_PASSWORD=change-this
   MINDBOTZ_API_TOKEN=optional-long-random-token
   CORRIDORKEY_RESOLUTION=1920x1080x24
   CORRIDORKEY_INSTALL_SAM2=y
   CORRIDORKEY_PREDOWNLOAD_SAM2=n
   CORRIDORKEY_INSTALL_GVM=n
   CORRIDORKEY_INSTALL_VIDEOMAMA=n
   MINDBOTZ_API_ENABLE_COMMANDS=0

6. Mount a RunPod network volume to /opt/corridorkey if you want models/projects/uploads to persist across pod rebuilds.

API Examples

curl http://localhost:8080/health
curl http://localhost:8080/status
curl http://localhost:8080/inputs
curl http://localhost:8080/outputs
curl http://localhost:8080/jobs

If MINDBOTZ_API_TOKEN is set:

curl -H "Authorization: Bearer $MINDBOTZ_API_TOKEN" http://localhost:8080/health

Record a job note:

curl -X POST http://localhost:8080/jobs \
  -H 'Content-Type: application/json' \
  -d '{"kind":"inference-note","clip":"example.mov","notes":"prepare for CorridorKey GUI run"}'

By default jobs are recorded but heavy commands are not executed. To permit command jobs, set MINDBOTZ_API_ENABLE_COMMANDS=1 and send a command field. Keep that disabled on public pods.

MCP Examples

HTTP JSON-RPC:

curl -X POST http://localhost:8765 \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

curl -X POST http://localhost:8765 \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"health","arguments":{}}}'

stdio MCP mode:

python -m mindbotz.mcp_server --transport stdio

Available tools:

• health
• status
• list_inputs
• list_outputs
• list_jobs
• create_job

What Changed From Upstream

See docs/mindbotz/MODIFICATIONS.md.

Security Notes

• Change MINDBOTZ_WORKSTATION_PASSWORD before exposing any endpoint.
• Set MINDBOTZ_API_TOKEN if API endpoints are reachable by anyone except you.
• Keep MINDBOTZ_API_ENABLE_COMMANDS=0 unless you explicitly need remote command execution.
• Raw VNC (5900) should generally stay private; use noVNC through RunPod's HTTP proxy instead.

Upstream Docs

The original upstream README is preserved as README_UPSTREAM.md.