Breaking

Open-Source Tool "Claude Video" Gives Claude Code Video Understanding

July 8, 2026 at 19:45 EDT

AI Agents
Software Dev & Coding
Open Source

An open-source tool called "Claude Video" has been released to let Anthropic's terminal-based coding agent "Claude Code" read the visual content of videos. It passes video frame images and transcripts to the agent through local processing, filling a gap where video handling was previously text-only.

Open-Source Tool · Claude Code Extension

Claude Code Just Got Eyes and Ears

A new open-source tool, Claude Video, lets Anthropic's terminal coding agent read the visual content of videos — passing frame images and transcripts to the agent through fully local processing, closing a gap that was previously text-only.

100%

local processing — no external cloud API

ingestion modes by goal & cost

inputs fused: frames + transcript

Four Modes — Frame Limit to Scale

Transcript

captions only

Efficient

by properties

100

Balanced

scene + keywords

∞

Token Burner

high fidelity

Frame limit per mode — height shown to scale (unlimited for Token Burner)

How a Video Becomes Context

Video In

YouTube URL or local .mov / recording

→

Local Process

Whisper transcript + FFmpeg scene-change frames

→

Into Claude

Frames + text passed end-to-end to the agent

What Developers Value

Turns client calls & consulting recordings into structured notes and action items
Makes notes searchable with zero manual work
Captures on-screen visual context, not just text

Limits to Keep in Mind

Token consumption rises with more frames
Initial local Whisper download required
Bound by Claude Code's context window
An extension via tooling — not fully native video understanding

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.