Scrape Everything
AI agents

Live social data for AI agents

Give your agent live public data from 9 platforms through MCP, one tool per endpoint, with the cost shown before it runs.

Get 100 free creditsRead the docs

Why it fits

When an agent has to answer what a creator posted this week, what a video says or how a brand account is growing, it needs a tool that fetches the public page now and returns structured fields. Scrape Everything provides that for Instagram, TikTok, X, YouTube, Facebook, Telegram, Spotify, Pinterest and Twitch, through an MCP server, a CLI and Agent Skills.

  • One MCP tool per API endpoint, plus get_account and list_scrapers, which are free. Each tool shows its credit cost.
  • Typed JSON with stable field names, and stable error codes such as NOT_FOUND, PRIVATE_ACCOUNT and RATE_LIMITED that a model can branch on.
  • Failed calls (blocks, timeouts, errors on our side) are refunded automatically.
  • Claude, Cursor, VS Code, Windsurf, Claude Code, Codex and Gemini CLI all use one key and one credit balance.
  • The key stays on your server: the API refuses browser requests, and each key can be locked to your server IPs.

In more detail

One tool per endpoint

The MCP server registers one tool for every endpoint in the catalog, plus get_account and list_scrapers, which cost nothing. Each tool carries its input schema and its credit cost, so the model can pick the right call and you can see what a step will spend before it runs. There is no generic "browse the web" tool that guesses at HTML: the agent asks for a TikTok transcript or an Instagram profile and gets typed JSON with stable field names.

Built for agent loops

Agents retry, branch and repeat themselves. Failed calls (blocks, timeouts, errors on our side) are refunded automatically, and every error carries a stable code such as NOT_FOUND, PRIVATE_ACCOUNT or RATE_LIMITED that a model can branch on instead of parsing a message. If the agent looks up the same profile twice in one session, add cache_days and the second read is free. Every response includes a request_id, which makes it easy to trace what an agent did after the fact.

Works where your agent runs

Claude, Cursor, VS Code and Windsurf connect through MCP with one config block. Claude Code, Codex and Gemini CLI can load the Agent Skills, which bundle workflows such as YouTube video research, comment sentiment mining and transcript notes. Scripted agents can shell out to the CLI and pipe JSON into jq. All of them use the same API key and the same credit balance, so moving a prototype from a chat window to a scheduled job does not change billing or data shape.

Keep the key on the server

The API refuses browser requests and rejects keys sent in the URL, so an agent that runs in a web front end must call through your own backend. For server-side agents, lock each key to your server IPs with an allowlist and keep it in the SCRAPEEVERYTHING_API_KEY environment variable that the MCP server, CLI and skills all read.

Live by default, cached when you allow it

Without cache_days, every tool call fetches the public page at the moment the agent asks, so answers reflect what is visible now. When an older result is good enough, for example a profile the agent already read earlier in the conversation, pass cache_days between 1 and 14 and the response comes from your private cache at 0 credits, marked cached: true with the time it was fetched. That lets the model state how fresh its evidence is instead of presenting stale numbers as current.

What it costs

50 research briefs a month, each from 1 YouTube search and 10 transcripts

  • YouTube search50 calls × 1 credit = 50
  • YouTube transcript500 calls × 1 credit = 500
  • Total550 credits

≈ $0.86 at the Builder rate ($1.56 per 1,000 credits; the pack is 25,000 credits for $39.00)

≈ $0.65 at the Growth rate ($1.19 per 1,000 credits; the pack is 100,000 credits for $119.00)

≈ $0.49 at the Scale rate ($0.90 per 1,000 credits; the pack is 500,000 credits for $449.00)

Packs are bought whole. Credits you do not use stay on your balance and never expire. Failed calls are refunded and not counted here. Credit calculator

Endpoints to use

Each endpoint has its own page with fields, limits and an example response.

Example: research a topic from YouTube

  1. The agent calls the YouTube search tool with a keyword and a small limit.
  2. It picks the most relevant videos from the titles and view counts.
  3. It fetches each transcript and writes a brief that cites timestamps.
  4. Repeat runs within your cache window reuse stored transcripts for free.
cURL
curl -X POST https://api.scrapeeverything.com/v1/youtube/search \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query":"lofi hip hop","limit":10}'

Integrations that fit

AI agents: questions

Which MCP clients are supported?

Any MCP client that runs a local stdio server, including Claude, Cursor, VS Code and Windsurf. Setup steps for each are on the MCP integration page.

Can the agent see what a call costs?

Yes. Each tool lists its credit cost, list_scrapers returns every endpoint with its cost, and get_account shows the remaining balance. Both cost 0 credits.

What happens when a call fails?

Blocks, timeouts and errors on our side are refunded automatically. The response carries a stable error code the agent can act on, plus a request_id.

Can I run the agent in the browser?

Not with the key in the browser. The API refuses browser requests, so route calls through your own server and keep the key there.

Start with 100 free credits.

Continue with Google, create an API key and send your first request in under a minute.

Get 100 free credits