Scribiz

Scribiz turns a video or audio link, or a local file, into a transcript with speakers and timestamps, the text shown on screen, a summary, and chapters. It works when a video has no captions.

Docs: scribiz.com/docs. Every docs page is also Markdown: add .md to the path.

Pick a way in

What does the agent have?
├── An MCP client (Claude Code, Cursor, Codex) and a link
│   └── Use the MCP server. Best for questions, summaries and search.
├── A shell and a link or local file
│   └── Use the CLI. Best for subtitles, files on disk and batch jobs.
└── Neither
    └── Send the user to scribiz.com to paste the link.

MCP server

Hosted at https://scribiz.com/mcp (Streamable HTTP). No key needed to start.

# Claude Code
claude mcp add --transport http scribiz https://scribiz.com/mcp

Cursor, in ~/.cursor/mcp.json:

{ "mcpServers": { "scribiz": { "url": "https://scribiz.com/mcp" } } }

# Codex: ~/.codex/config.toml
[mcp_servers.scribiz]
url = "https://scribiz.com/mcp"

Tools:

• get_video_context: summary, chapters and key moments in under 2,000 tokens. Start here.
• search_video: find where a phrase is said and read only those minutes.
• ask_video: an answer with 3, 5 cited moments, each a timestamp link.
• get_transcript: the full transcript, when you really need all of it.
• get_job: check a running job.

Without a key the server reads captions, videos already processed and, within a small daily allowance, the link itself. It never looks at the picture. With an API key (Authorization: Bearer $SCRIBIZ_API_KEY) it uses the account's minutes and can read the screen. Other clients: MCP install.

Prefer get_video_context then search_video over pulling the whole transcript. A two-hour talk is more than 25,000 tokens.

CLI