The social data layer for AI agents

Why agents like this API

• Transcripts and AI summaries are first-class endpoints, not an afterthought. They're the exact inputs an LLM wants.
• Consistent fields across platforms: one tool definition covers five sites.
• No HTML soup. Every response is compact JSON with predictable keys, so you spend tokens on content instead of markup.
• MCP server at mcp.scrapersocial.com exposes every endpoint as a tool for Claude and any MCP client.
• Credit headers on every response (X-Credits-Charged, X-Credits-Remaining) let an agent budget itself mid-run instead of discovering an empty balance at the end.

A research agent in one prompt

Connect the MCP server, then ask in plain language. The agent chains get_channel_posts and get_transcript on its own:

> pull the last 10 reels from @thefoodlab,
  transcribe them, and rank by views

tool: get_channel_reels(handle="thefoodlab", platform="instagram")
tool: get_transcript(url=…) x 10
result: ranked table with views, likes, and a one-line
        summary per reel

What that run costs, so you can budget it: ten reels from channel-reels is 30 credits, ten Instagram transcripts at 40 credits each is 400, so about 430 credits total, roughly $2.15 at the monthly rate. The same workflow on TikTok is much cheaper: 110 credits, about $0.55, because TikTok transcripts usually come from captions rather than speech-to-text. Rerunning the pipeline later hits the cache and returns in milliseconds.

Or wire it up as a function tool

Any framework that speaks function calling can wrap an endpoint in a few lines:

const tools = [{
  name: "get_tiktok_transcript",
  description: "Transcript of a public TikTok video",
  input_schema: { type: "object", properties: { url: { type: "string" } } },
  run: async ({ url }) =>
    (await fetch(`https://api.scrapersocial.com/v1/tiktok/transcript?url=${url}`, {
      headers: { Authorization: `Bearer ${process.env.SCRAPERSOCIAL_KEY}` },
    })).json(),
}];

Built to be found by agents, not just used by them

A growing share of our signups start with an agent, not a person. So the machine-readable surface is a product feature: an OpenAPI 3 spec at /openapi.json, llms.txt and a full docs mirror in llms-full.txt, plus .well-known discovery endpoints: an API catalog, the MCP server card, and agent skills. Docs pages advertise all of these in RFC 8288 Link headers. If your agent can read HTTP, it can find everything without a human in the loop.

Full setup guides for Claude, Claude Code, Cursor, and ChatGPT live on the MCP page, and agent-flavored docs at /ai-agents.