Your agent can't watch a video. So we turn it into text.
Give vid2llm a TikTok, YouTube, Instagram or X link. It returns who posted it, the description, and a timestamped transcript, as plain text your agent can work with. Agents pay per video with x402, from one cent. No account, no API key.
Add it to Claude Code
claude mcp add vid2llm \
-e VID2LLM_URL=https://vid2llm.dev \
-e VID2LLM_PRIVATE_KEY=0x… \
-- node /path/to/vid2llm/mcp/dist/index.js
This adds a watch_video tool, and tells your agent what it is for. It pays
from the wallet whose key you give it, never more than the limit you set.
What your agent gets back
One markdown document per video. This is the real response for the first video ever uploaded to YouTube.
# Me at the zoo
- **Platform:** YouTube
- **URL:** https://www.youtube.com/watch?v=jNQXAC9IVRw
- **Author:** jawed
- **Author profile:** https://www.youtube.com/@jawed
- **Uploaded:** 2005-04-24
- **Duration:** 0:19
- **Views:** 442,573,417
- **Likes:** 20,047,934
- **Tags:** me at the zoo, jawed karim, first youtube video
- **Spoken language:** en
## Description
00:00 Intro
00:05 The cool thing
00:17 End
## Transcript
[0:00] Alright, so here we are, one of the elephants.
[0:05] The cool thing about these guys is that they have really, really, really long trunks.
[0:13] And that's cool.
[0:17] And that's pretty much all there is to say.
What it costs
The price follows the video's length, billed per started minute. Your agent sees the exact amount before it pays, and pays nothing if the video can't be read.
Audio only $0.01
With visuals $0.07
How it works without the MCP server
Any agent with an x402 client can call it directly. Put the video link after the domain.
https://vid2llm.dev/https://www.tiktok.com/@user/video/123
-
Your agent asks for the video
It gets
402 Payment Requiredback, with the exact price for that video in USDC. -
It pays and asks again
An x402 client signs the payment and repeats the request. No sign-up happens anywhere.
-
It reads the result
The response is the markdown document above. Add
?visuals=1for on-screen text and what is shown, or?format=jsonfor JSON.
What it can read
- Public videos on TikTok, YouTube, Instagram and X, up to 15 minutes.
- Other public pages and direct video file links, when the site's robots.txt allows it.
- Nothing behind a login.
Questions
- How can an AI agent watch a TikTok or YouTube video?
- It can't watch one directly. vid2llm downloads the public video, transcribes the audio and returns the author, description, stats and a timestamped transcript as text, which the agent reads instead.
- How do I let Claude Code read videos?
- Add the vid2llm MCP server with
claude mcp add. It gives Claude awatch_videotool, tells it when to use it, and pays per video from a wallet you fund, up to a limit you set. - Do I need an account or an API key?
- No. Requests are paid individually with the x402 protocol in USDC. An unpaid request gets a
402 Payment Requiredresponse naming the exact price, and the agent pays and repeats it. - What does it cost to transcribe a video?
- The price follows the video's length, billed per started minute, starting at one cent for a clip under a minute. Describing what is shown on screen costs extra. Nothing is charged if the video can't be read.
- Which sites does vid2llm support?
- Public videos on TikTok, YouTube including Shorts, Instagram Reels and X, plus other public pages and direct video file links when the site's robots.txt allows it. It does not read anything behind a login.