Boson-Video

Get the point of any video without watching all of it. Paste a YouTube link or drop a video
file. Every line of the summary, the transcript and the screen text carries the second it came
from, and questions are answered from the video and checked against it.

• The ribbon: the whole video on one strip (scenes, chapters, most replayed). Hover to see any second's frame and words.
• Summary: sections with their frames; every sentence timed and checked (✓ ? ✗).
• Transcript: the original with English beneath, technical terms explained, on-screen text read in.
• Ask: answers that cite their moments and show the frames, with background kept apart.
• Your own AI: the same document as an MCP plugin for Claude, Cursor, Codex and more.

Built for long talks, lectures and finance videos in a language you half know (Chinese and English today).

In your own AI

One line, nothing else to install (uv runs it):

claude mcp add boson-video -- uvx boson-video mcp

Claude Desktop (claude_desktop_config.json) or Cursor (.cursor/mcp.json):

{ "mcpServers": { "boson-video": { "command": "uvx", "args": ["boson-video", "mcp"] } } }

Codex (~/.codex/config.toml):

[mcp_servers.boson-video]
command = "uvx"
args = ["boson-video", "mcp"]

Then ask your AI about any YouTube link. It opens the video, reads the transcript, looks at the
frames that matter at full resolution, searches, and checks its claims, citing every moment. No
keys needed. Everything runs on your computer. In apps that show interactive pages (Claude Desktop,
claude.ai, ChatGPT, Cursor), the video's page opens right in the chat: the player, the ribbon, the
transcript with English beneath, terms and scenes, and questions you ask there go to your AI. Details: docs/plugin.md.

Run the page

uvx boson-video web             # http://127.0.0.1:8770; the first run prints your invite code

Keys make it fuller: INCEPTION_API_KEY (Mercury writes the summary, the English and the terms)
and TYPESAFE_API_KEY (Jev checks every sentence and answers questions). Put them in a .env file in
the folder you start from, or in ~/.boson-video/.env; the server warns at start when one is missing.
Without them you still get the scenes, the transcript and search. Any OpenAI-compatible model can
write instead of Mercury: set BOSON_WRITER_BASE_URL, BOSON_WRITER_MODEL and BOSON_WRITER_API_KEY.
Optional and off by default: with OPENAI_API_KEY and BOSON_SCREEN_CHECK=1, a sentence the words can't
confirm gets a second look at the frames on screen (OpenAI's Decisions API, about a cent a video).

Tested videos

Measured, one run each unless a range is shown. Mac: M5 MacBook with Apple's transcriber. Windows: a 12-thread laptop with
SenseVoice or Parakeet on the CPU. Full tables: docs/measured.md.

¹ Against human captions (scripts/accuracy.py); about half the English errors are filler words the captions leave out.
Summary ✓: sentences confirmed against what was said (Jev and code).

Command line

uvx boson-video <link | id | file>        # build a video: scenes, words, screen text, summary
uvx boson-video ask <video> "question"    # the moment that answers it
uvx boson-video models                    # fetch the local speech models now instead of on first use

From a clone: uv sync, then uv run boson-video …; uv run pytest runs the tests (offline).

More

• docs/workflow.md: how the workflow is designed: stages, the document, the ways in
• docs/how-it-works.md: the pipeline, what YouTube allows, limits
• docs/DIRECTION.md: what we're building, milestones, decisions
• docs/timeline-format.md: the document format, for other tools
• AGENTS.md: the guide for coding agents

MIT licensed. YouTube's terms don't allow automated access for products, so everything that touches
YouTube runs on your own computer.