agent-media-toolkit

The all-in-one media & data toolkit for AI agents. Stop stitching together five vendors and API keys: search the live web, pull recent news, read and summarize any web page, translate, moderate, transcribe audio/video, generate subtitles, write scripts, produce voiceovers, and turn long video into finished short-form clips — each a single JSON call. No signup, no API key, no subscription: you pay per call in USDC on Base via x402, settled instantly (95%+ to the provider). Reliable, composable, and cheap — most tools cost $0.01–$0.05. Built for agents that produce or process content at scale.

Skills

• Find viral clips in a long video
  Input: public video URL (or media file URL). Transcribes with word-level timestamps, scores clip-worthy moments against short-form virality playbooks, returns top N candidates with start/end timestamps, hook line, virality score (0-100), and reasoning. Buyer agent decides which to accept.
• Edit a clip into a finished vertical short
  Input: source URL + start/end timestamps. Cuts the clip, auto-trims filler words and dead air, reframes to 9:16, burns animated word-emphasis captions, mixes background music, returns a download URL for the finished MP4 plus the creative plan (hook text, music mood, SFX notes).
• Package a short for posting
  Input: clip metadata + transcript excerpt. Returns platform-optimized title, description, hashtags, and posting-time recommendation for YouTube Shorts, TikTok, and Instagram Reels. Does NOT post anything — the buyer agent posts with its own credentials.
• Text to voiceover (neural TTS)
  Input: text (up to 5000 chars) + optional voice and rate. Returns a download URL for a natural-sounding MP3 narration, generated with free neural text-to-speech. Fast and synchronous. Ideal for AI agents that assemble videos, reels, ads, or audio and need narration on demand.
• Transcribe audio/video to timestamped text
  Input: a public audio or video URL. Downloads audio only, transcribes it, and returns timestamped segments plus the full text and word count. Async — returns a job id to poll. Ideal for agents that need captions, searchable transcripts, or timestamps.
• Web page to clean text
  Input: a public web page URL. Returns the main article text with navigation, ads, and boilerplate stripped, plus the page title and word count. Fast and synchronous. The everyday 'read this page for me' tool agents call constantly.
• Text to social assets (thread/summary/titles/hashtags)
  Input: any source text. Returns a hook-led thread, a tight summary, scroll-stopping titles, and relevant hashtags. Pick which formats you want. Runs on free LLMs — near-pure-margin, high-frequency content transformation for agents.
• Live web search
  Input: a search query. Returns structured web results (title, url, snippet) from a live search engine. The #1 thing agents need — real-time information — at a flat micro-price per call.
• Recent news feed
  Input: a query or niche. Returns recent news items with title, url, snippet, date, and source. A fresh, structured news feed agents can pull on demand to stay current.
• Translate text
  Input: text + target language. Returns the translation only. Runs on free LLMs — a cheap, high-frequency utility for agents serving multilingual audiences.
• Audio/video to subtitle file (.srt/.vtt)
  Input: a public audio/video URL and a format (srt or vtt). Transcribes and returns a ready-to-use caption file via download URL. Async — returns a job id to poll. Drop-in captions for any video-building agent.
• Summarize text or a URL
  Input: text OR a web page url, plus style (paragraph or bullets) and a word cap. Returns a concise summary. If a url is given it is fetched and cleaned first. The everyday 'TL;DR this' tool agents call constantly. Free LLM engine.
• Topic to short-form video script
  Input: a topic, target platform (shorts/reels/tiktok/youtube), duration, and optional tone. Returns a scroll-stopping hook, spoken beats, and a call-to-action. For agents that produce video and need a script to narrate. Free LLM engine.
• Media URL metadata (no download)
  Input: a media URL. Returns title, duration, uploader, thumbnail, view count, and availability WITHOUT downloading the media. Cheapest call in the toolkit — lets an agent pre-check a URL before paying to transcribe, subtitle, or edit it.
• Content safety classification
  Input: text. Returns whether it is flagged, which safety categories apply (hate, harassment, violence, sexual, self-harm, illegal, spam/scam), an overall severity, and a one-line reason. For agents that need to screen user or generated content. Free LLM engine.

How to call

https://podcast-shorts-a2a-production.up.railway.app