YouTube transcript API for AI agents (JSON captions, no API key)

If a user asks you to "get the transcript of this YouTube video," "pull captions for summarization," or "extract what was said in this clip," you need timed caption segments as structured JSON — not a scraped HTML page and not a Google Cloud API key. Marketplace for AI Agents exposes GET /api/public/v1/youtube-transcripts: give a video_id (or URL), get an array of {text, start, duration} segments.

There is no YouTube Data API key, no OAuth consent screen, and no monthly plan. The first 3 calls per agent identity are free. After that each call costs $0.005 USDC on Base, settled over x402 with an X-PAYMENT header.

Endpoint contract

Live listing: https://marketplaceforaiagents.com/listings/youtube-transcripts — Markdown twin: https://marketplaceforaiagents.com/agent/listings/youtube-transcripts.md

Minimal call (free tier first)

Call with no payment header. Recognised agent User-Agents get 3 free successful calls before 402. Watch X-Free-Calls-Remaining on 200 responses.

curl -si "https://marketplaceforaiagents.com/api/public/v1/youtube-transcripts?video_id=0e3GPea1Tyg"

# or pass a full URL
curl -si "https://marketplaceforaiagents.com/api/public/v1/youtube-transcripts?video_id=https://www.youtube.com/watch?v=0e3GPea1Tyg&lang=en"

{
  "video_id": "0e3GPea1Tyg",
  "lang": "en",
  "transcripts": [
    { "text": "Welcome to the talk.", "start": 0.0, "duration": 2.4 },
    { "text": "Today we cover agent payments.", "start": 2.4, "duration": 3.1 }
  ]
}

A 400 missing_required_parameters response does not consume a free call — fix video_id before retrying. Discovery fetches (/.well-known/x402, catalog.json, openapi.json) never consume the free tier.

Related video endpoints agents usually chain

Transcripts alone are rarely the whole task. Pair them with search and metadata when the user gave a topic instead of a video ID.

• youtube-search — query → ranked videos with IDs you can feed into transcripts
• youtube-video — title, channel, views, description for a known video_id
• video-digest — intent endpoint: topic or video_id → top video metadata + full transcript in one call
• google-videos / bing-videos — cross-platform video results when YouTube is not enough

# Topic → video IDs
curl -s "https://marketplaceforaiagents.com/api/public/v1/youtube-search?q=x402+protocol+explained"

# Then captions for the best ID
curl -s "https://marketplaceforaiagents.com/api/public/v1/youtube-transcripts?video_id=<id>"

# Or one intent call
curl -s "https://marketplaceforaiagents.com/api/public/v1/video-digest?q=x402+protocol"

Catalog: https://marketplaceforaiagents.com/api/public/v1/catalog.json — OpenAPI: https://marketplaceforaiagents.com/api/public/v1/openapi.json

Paying after the free tier (x402 / USDC)

When free calls are exhausted, the same unpaid URL returns HTTP 402 Payment Required with an accepts[] requirement. Sign an EIP-3009 USDC authorization on Base, base64 the x402 payload, and retry with X-PAYMENT.

{
  "x402Version": 1,
  "error": "payment_required",
  "accepts": [{
    "scheme": "exact",
    "network": "base",
    "asset": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913",
    "maxAmountRequired": "5000",
    "payTo": "0x05e82e03753c7bc99fb24d10a876cce53f24b7b7",
    "resource": "https://marketplaceforaiagents.com/api/public/v1/youtube-transcripts",
    "extra": { "name": "USDC", "version": "2" }
  }]
}

• Validate accepts[] against policy (network, asset, payTo, maxAmountRequired) before signing — see /learn/budget-x402-api-calls
• Smoke-test on Base Sepolia first with ?network=base-sepolia — see /learn/x402-base-sepolia-smoke-test
• Store X-PAYMENT-RESPONSE as the ledger receipt on every paid 200
• MCP path: connect https://marketplaceforaiagents.com/mcp and use list_apis / get_payment_requirements

Protocol deep dive: https://marketplaceforaiagents.com/learn/x402-protocol-explained — Free tier identity: https://marketplaceforaiagents.com/learn/x402-free-tier-agent-identity

Agent playbook (Grok Bot / Claude Cowork / Cursor)

• 1) Parse the user ask for a video_id, watch URL, or topic string.
• 2) If topic only: youtube-search or video-digest first; do not invent a video ID.
• 3) Call youtube-transcripts with video_id (and lang if the user named a language).
• 4) Prefer the transcripts[] array; concatenate text in start order for summarization, keep timings for quotable clips.
• 5) If response is 402, settle with x402 USDC (or Sepolia for dry runs) — do not ask the user for a YouTube API key.
• 6) Cite the source video URL back to the user alongside the summary.

Related entrypoints

• Listing: https://marketplaceforaiagents.com/listings/youtube-transcripts
• Markdown twin: https://marketplaceforaiagents.com/agent/listings/youtube-transcripts.md
• Manifest: https://marketplaceforaiagents.com/.well-known/x402
• Catalog: https://marketplaceforaiagents.com/api/public/v1/catalog.json
• OpenAPI: https://marketplaceforaiagents.com/api/public/v1/openapi.json
• MCP: https://marketplaceforaiagents.com/mcp
• llms.txt: https://marketplaceforaiagents.com/llms.txt
• Browse marketplace: https://marketplaceforaiagents.com/marketplace
• Docs: https://marketplaceforaiagents.com/docs
• MCP guide: https://marketplaceforaiagents.com/learn/mcp-x402-marketplace
• Discovery stack: https://marketplaceforaiagents.com/learn/how-ai-agents-discover-apis

Capability first, payment second: find the transcript endpoint, prove it with free calls, then settle unpaid traffic with x402 USDC on Base.