How huangchihhungleo/claude-real-video works
claude-real-video is a local video-analysis tool that turns a URL or file into timestamped keyframes and a transcript for an LLM, with CLI, local web, and MCP entry points. The main flow is source acquisition, windowed scene-aware frame selection and deduplication, optional transcription and speaker labeling, then browsable or machine-readable outputs. A separate recall flow indexes analyses locally for later cross-video search. Core internals are represented at the capability level; the sampled core.py excerpt is partial, so uncertain implementation wiring is omitted.
Main entry points: core.py Analysis pipeline · mcp_server.py MCP server · cli.py CLI · memory.py Memory index
Components 19
User interfaces
- CLI
src/claude_real_video/cli.py - Local web UI
src/claude_real_video/serve.py - MCP server
src/claude_real_video/mcp_server.py
Video analysis
- Analysis pipeline
src/claude_real_video/core.py - Time window
src/claude_real_video/core.py - Scene frames
src/claude_real_video/core.py - Transcription
src/claude_real_video/core.py - Speaker labels
src/claude_real_video/speakers.py - Temporal check
src/claude_real_video/temporal_check.py
Analysis outputs
- Output artifacts
src/claude_real_video/core.py - Local viewer
src/claude_real_video/viewer.py
Persistent recall
- Memory index
src/claude_real_video/memory.py - Cross-video search
src/claude_real_video/memory.py
Other components
- User
- LLM agent
- URL or file
- ffmpeg
- Whisper backend
- sherpa-onnx
Connections 24
- User → CLI: runs analysis
- User → Local web UI: submits video
- LLM agent → MCP server: requests analysis
- CLI → Analysis pipeline: calls pipeline
- Local web UI → CLI: launches command
- MCP server → Analysis pipeline: calls pipeline
- URL or file → Analysis pipeline: provides video
- Analysis pipeline → Time window: applies range
- Time window → Scene frames: selects keyframes
- Analysis pipeline → ffmpeg: extracts video data
- Analysis pipeline → Transcription: transcribes audio
- Transcription → Whisper backend: uses backend
- Transcription → Speaker labels: adds speaker labels
- Analysis pipeline → Temporal check: checks timing
- Analysis pipeline → Output artifacts: writes results
- CLI → Local viewer: generates viewer
- Output artifacts → Local viewer: reads frames and transcript
- CLI → Memory index: indexes analysis
- MCP server → Memory index: indexes analysis
- Output artifacts → Memory index: reads manifest and transcript
- User → Cross-video search: searches videos
- Cross-video search → Memory index: queries index
- MCP server → Cross-video search: searches memory
- MCP server → LLM agent: returns frames and transcript