Skip to content
ScraperSocial API

For AI agents

ScraperSocial is built to be consumed by agents: transcripts and summaries are first-class, responses are compact JSON with stable keys, and every capability is exposed as a tool.

Option 1 — MCP server (recommended)

One config block gives any MCP client the full surface:

{
  "mcpServers": {
    "scrapersocial": {
      "url": "https://mcp.scrapersocial.com",
      "headers": { "Authorization": "Bearer sk_live_..." }
    }
  }
}

OAuth 2.1 (PKCE) is also supported — clients like claude.ai can connect with no key at all. In Claude Code:

Terminal window
claude mcp add --transport http scrapersocial https://mcp.scrapersocial.com

Tools exposed

The server exposes 14 tools. The capability tools cover the common reads; list_endpoints and call_endpoint reach every other endpoint.

  • get_transcript — Spoken-word transcript of a public video, with timestamped segments where the source publishes them.
  • get_summary — Short LLM summary of what a public video actually says, built on its transcript.
  • get_post_stats — Engagement stats for one public post or video: views, likes, comments, shares and reposts where the platform publishes them.
  • get_comments — Comments on a public post or video, newest first, paginated with limit and cursor.
  • get_comment_replies — The reply thread hanging off a single public comment.
  • get_profile — Profile, page or channel record for a public account: follower and subscriber counts, bio, and totals.
  • get_channel_posts — Recent public posts or videos from one account, with per-item metrics.
  • search — Keyword search over one platform’s public content.
  • search_hashtag — Public posts carrying a hashtag.
  • get_reviews — Public reviews and ratings for a product, business, app or company page.
  • download_media — Direct media URLs for a public post or video, served from the platform’s own CDN.
  • get_trending — What is currently trending or rising on a platform.
  • list_endpoints — Browse the whole API surface — every endpoint, its input and its credit cost. Free, and calls nothing upstream.
  • call_endpoint — Run any endpoint by its path, so nothing on the API is out of reach of an agent.

Option 2 — function-calling tools

Wrap any endpoint in your framework of choice:

def get_tiktok_transcript(url: str) -> dict:
    """Transcript of a public TikTok video, with timestamped segments."""
    r = httpx.get("https://api.scrapersocial.com/v1/tiktok/transcript",
                  params={"url": url},
                  headers={"Authorization": f"Bearer {KEY}"}, timeout=120)
    r.raise_for_status()
    return r.json()["data"]

Tips that make agents behave:

  • Use format=text for transcripts headed straight into a prompt.
  • Use fields= projections to keep tool results token-cheap.
  • Surface error.message to the model — messages are written to be self-correcting (“It may be private or deleted”).

Discovery endpoints

Agents can find everything without a human:

Next: Recipes · MCP marketing page