socialcrawl/mcp

---

Installation

Remote server (hosted — no install)

Connect straight to the hosted Streamable HTTP endpoint — nothing to install or run:

Claude Code (works in the CLI and Claude Code on the web / cloud sandboxes)

claude mcp add --scope user --transport http socialcrawl https://mcp.socialcrawl.dev/mcp \
  --header "Authorization: Bearer sc_your_key_here"

Any client that reads .mcp.json

{
  "mcpServers": {
    "socialcrawl": {
      "type": "http",
      "url": "https://mcp.socialcrawl.dev/mcp",
      "headers": { "Authorization": "Bearer ${SOCIALCRAWL_API_KEY}" }
    }
  }
}

Cursor / Windsurf / VS Code — choose the HTTP (“streamable-http”) server type with the same URL and header. x-api-key: sc_your_key_here works as an alternative header.

The discovery tools (socialcrawl_find, socialcrawl_endpoint, socialcrawl_estimate) work without a key, so you can explore before signing up. claude.ai custom connectors (Settings → Connectors) require OAuth. An OAuth 2.1 resource-server mode is in preview and off by default (SOCIALCRAWL_OAUTH=1; see docs/REMOTE-STREAMABLE-HTTP.md); until it is enabled on the hosted server, use the header-based setup above.

Prefer running it locally? Every stdio option below works exactly as before.

npm

npm install -g socialcrawl-mcp

Available on npm. Most users don’t need this — the MCP client configs below use npx and auto-install on first run.

Claude Code (quickest)

claude mcp add --scope user socialcrawl -e SOCIALCRAWL_API_KEY=sc_your_key_here -- npx -y socialcrawl-mcp

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "socialcrawl": {
      "command": "npx",
      "args": ["-y", "socialcrawl-mcp"],
      "env": {
        "SOCIALCRAWL_API_KEY": "sc_your_key_here"
      }
    }
  }
}

Cursor

Add to .cursor/mcp.json in your project root or ~/.cursor/mcp.json globally:

{
  "mcpServers": {
    "socialcrawl": {
      "command": "npx",
      "args": ["-y", "socialcrawl-mcp"],
      "env": {
        "SOCIALCRAWL_API_KEY": "sc_your_key_here"
      }
    }
  }
}

VS Code (Claude Code)

Add to .vscode/mcp.json in your project or your user settings:

{
  "servers": {
    "socialcrawl": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "socialcrawl-mcp"],
      "env": {
        "SOCIALCRAWL_API_KEY": "sc_your_key_here"
      }
    }
  }
}

Windsurf

Add to your Windsurf MCP configuration:

{
  "mcpServers": {
    "socialcrawl": {
      "command": "npx",
      "args": ["-y", "socialcrawl-mcp"],
      "env": {
        "SOCIALCRAWL_API_KEY": "sc_your_key_here"
      }
    }
  }
}

Other MCP-compatible clients

Any MCP client that supports stdio transport can use this server. The general pattern is:

• Command: npx
• Args: ["-y", "socialcrawl-mcp"]
• Environment: SOCIALCRAWL_API_KEY set to your API key

Restart your AI client after saving the configuration.

Overview

socialcrawl-mcp is an MCP (Model Context Protocol) server that connects AI agents to the SocialCrawl API — a unified data API over social, commerce, jobs, finance, research and web data. socialcrawl_find with no task lists every platform; socialcrawl_account view freshness shows the live counts.

Retrieve profiles, posts, comments, search results, trending content, and analytics from TikTok, Instagram, YouTube, Twitter/X, LinkedIn, Reddit, Threads, Douyin, Xiaohongshu, Telegram, Quora, Bluesky, Apple Music, GitHub, Hacker News, Product Hunt, Polymarket, and many more platforms. Pull products, offers, price history, reviews, and sellers from Amazon, Walmart, Target, Home Depot, eBay, Klarna, AliExpress, Etsy, Sephora, H&M, Kohl’s, Wayfair, Gumtree, and Google Shopping; apps, charts, and reviews from Google Play and the Apple App Store; hotels, restaurants, attractions, cruises, and traveler reviews from Tripadvisor, plus local business data from Yelp and Google Business; brand reputation from Trustpilot and software reviews from G2; job listings and salary bands from LinkedIn, Indeed, Bing, and Xing; market quotes, price history, financial statements, and options chains; US congressional trading disclosures; Korean search and Data Lab trend series from Naver; cross-web brand mentions with sentiment via Content Analysis; Google News headlines and Google Trends interest curves — plus web research via Tavily and Perplexity, AI-powered X search via Grok, on-page SEO audits, and a single /search/everywhere endpoint that fans out across 14 sources in one call.

New in v2.0.0: seven tools instead of eleven (socialcrawl_find, socialcrawl_endpoint, socialcrawl_estimate, socialcrawl_request, socialcrawl_collect, socialcrawl_account, socialcrawl_manage). See Available Tools for the 1.x → 2.0 mapping and CHANGELOG.md for every release.

What the MCP server does:

• Finds the right endpoint for a task in plain words (ranked search across every endpoint, with URLs and handles resolved), and tells the agent what data each one returns
• Fetches live data on your behalf across every platform, method, and composite
• Prices every call up front — ladder, flat, or metered band — so an agent can budget before it spends
• Validates requests locally before calling the API: required params, oneOf groups, enum values, integer ranges, parameter couplings, and CSV limits. A bad call fails free instead of burning credits
• Provides built-in API documentation the agent can query on demand, paged rather than truncated

Setup

1. Get your API key

Sign up at socialcrawl.dev and grab your API key from the dashboard. Every account starts with 100 free credits — no credit card required.

2. Add the key to your config

Replace sc_your_key_here in the installation config above with your actual API key (starts with sc_).

[!TIP] You can also set SOCIALCRAWL_API_KEY as a system environment variable instead of putting it in the MCP config. The discovery and documentation tools work even without a key — only actual API requests need one.

Usage

Ask your AI agent in natural language. The MCP server handles the rest.

Fetch a profile

Get the TikTok profile for @charlidamelio

The agent calls socialcrawl_request with platform: "tiktok", resource: "profile", params: { handle: "charlidamelio" } and returns structured profile data including followers, bio, verification status, and engagement metrics.

Search across platforms

Search YouTube for "machine learning tutorials"

Get post comments

Get the comments on this Instagram post: https://instagram.com/p/CwA1234abcd

Cross-platform research

Compare the follower counts of @mkbhd on TikTok, Instagram, YouTube, and Twitter

The agent makes 4 sequential API calls — one per platform — and compiles the results into a comparison.

Shop across retailers

Find the cheapest 65-inch OLED TV across Amazon, Walmart, Target, and eBay

Explore available endpoints

What social media platforms can you access?

Show me all the TikTok endpoints

Which endpoints can give me video transcripts?

Each is a socialcrawl_find call: the task in plain words, ranked across every endpoint, no platform needed.

Learn the API from the API

How do I get started with SocialCrawl?

Show me exactly how to call the Prism comments endpoint

Is this MCP server's endpoint list up to date?

These hit socialcrawl_endpoint and socialcrawl_account and cost nothing.

Check what something costs before running it

What would it cost to run a Prism brand-mentions report?

Show me everything I can call for 1 credit

Why did that last call charge me 7 credits instead of 2?

The first two hit socialcrawl_estimate; the third reads the credit ledger via socialcrawl_account with view: "transactions" and shows the deduction and refund rows for that request_id.

Access documentation

How does the SocialCrawl credit system work?

How do I page through a list endpoint?

Example response

Every response follows a unified envelope format:

{
  "success": true,
  "platform": "tiktok",
  "endpoint": "/v1/tiktok/profile",
  "data": {
    "content": { "text": "...", "media_urls": ["..."] },
    "author": { "username": "charlidamelio", "followers": 156000000 },
    "engagement": { "likes": 5200, "engagement_rate": 0.045 },
    "metadata": { "language": "en", "content_category": "entertainment" }
  },
  "credits_used": 1,
  "credits_remaining": 99
}

[!NOTE] The same response structure is returned for every platform — no per-platform parsing logic needed.

Available Tools

Version 2.0.0 exposes 7 tools (about 4.7k tokens of tools/list, down from about 12.8k for the 11 tools of 1.x):

Upgrading from 1.x. Set SOCIALCRAWL_LEGACY_TOOLS=1 to also register the nine retired 1.x names as thin wrappers for this major version. The mapping:

Discovery, live and bundled

socialcrawl_find, socialcrawl_endpoint and socialcrawl_estimate answer from data bundled with the server (generated from the backend registry, dump schema v4) and need no key. With a key they first ask the API’s free /v1/utility/* routes (find, resolve, endpoint, estimate), which answer from the live registry, and fall back to the bundled answer when a route is not deployed. Every one of these is 0 credits.

Why freshness matters. This server ships a catalogue generated when it was built; the API keeps moving. Data calls always hit the live API and keep working, but discovery, pricing and local validation answer from that snapshot. The server checks this itself once per process and, when it is behind, adds one line to the next tool result. One free call tells you which situation you are in:

socialcrawl_account  view: "freshness"

These same endpoints are plain HTTP, so a third-party integration or a non-MCP agent framework gets the identical information with a curl.

Pricing — know the cost before you spend

socialcrawl_estimate exists because a single number is a lie for most of the surface. SocialCrawl bills three ways:

• Ladder — the tier rate per request: standard 1cr, advanced 5cr, premium 10cr.
• Flat (some of them free) — a per-endpoint override, e.g. /v1/search/everywhere at 20cr flat.
• Metered — the charge depends on the request. An upfront ceiling is deducted and refunded down to the work actually done.

Quoting a metered endpoint’s base cost understates every call: /v1/search/news has a 1cr base but really charges 2–62cr — one credit per Google country leg that returns articles, plus per-article billing when the bing engine is added. The tool returns the real band, the registry’s own charging rule, the parameters that move the bill, and the worst case to budget for:

action: "overview"   → ladder, every free endpoint, every flat override, every metered band + rule, cache TTLs, full refund matrix
action: "endpoint"   → one endpoint's price, rule, price-driving params, paging cost, worst case
action: "platform"   → a whole platform's cost table
action: "list"       → rank/filter by cost — "everything I can call for 1 credit", "the 10 most expensive endpoints"
action: "hydration"  → every include= row join: what it fills, per-row rate, row cap, fully-joined ceiling
action: "judgments"  → every judged list: free default labels/relevance, metered presets, the hold per opt-in

Pass the include you intend to send to action: "endpoint" and the band becomes arithmetic — the exact hold, itemised per join, using the same formula the backend’s pricer runs:

action: "endpoint", platform: "linkedin", resource: "search/people", include: "profile", rows: 3
→ holds 22 credits (10cr page + 12cr for 3 rows of `profile`), settling anywhere down to 10

Or pass the exact params you will send (and calls, for a whole job). Joins, metered judgments and max_pages are all itemised, and max_pages multiplies the page, because each page walked is billed as one call:

action: "endpoint", platform: "youtube", resource: "search", calls: 10,
params: { query: "espresso", include: "engagement,channel", label: "mention", brand: "Acme", max_pages: "3" }
→ 1cr page + 5cr engagement + 5cr channel + 4cr label=mention = up to 15cr per page
  × 3 pages = up to 45cr per call; 10 calls = up to 450 credits held, settling to what was filled and judged

It also states the rules that make the real charge differ from the sticker price. Cache hits, idempotent replays, empty results, upstream failures and dry_run=1 previews are all 0 credits; seen= discounts each page by its share of rows you already received; since / stop_at_id end a walk at rows you already hold.

Judgments — labels and relevance, free by default

About 50 list endpoints judge their rows for you at no extra credit. Posts carry sponsored, intent and niche; comments carry sentiment, question, purchase_intent and complaint; reviews carry sentiment and issue. All of these land under computed.labels, and the judged searches score each row’s relevance to your query under computed.relevance. Only these add credits:

• a metered preset: mention (with brand=), quality, injection on posts; spam, toxic, low_quality on comments; reports, incentivized on reviews,
• label=intent with offer= (what you sell),
• relevance=score|filter with your own topic in relevant_to=.

Each holds 1 credit per started 25 rows of the page (4cr on a 100-row page) and settles to the rows judged fresh. Already-judged rows and cached pages are free. judgments=off turns the defaults off, and dry_run=1 previews the cost for 0 credits. socialcrawl_endpoint with id: "judgments" is the full contract.

Row hydration — one call instead of a page plus a lookup per row

Some lists are thin because the upstream publishes nothing else on them: a Pinterest search result has no save count, a LinkedIn reactor row no follower count, a YouTube playlist no view counts or durations. Another SocialCrawl endpoint answers each of those for one row — so those endpoints accept an include= token that joins every row to that sibling in the same call.

GET /v1/pinterest/search?query=kitchen&include=engagement      → saves, likes, comments, shares on every row
GET /v1/youtube/playlist?playlistId=PL...&include=engagement,channel
GET /v1/linkedin/search/people?keywords=cto&include=profile&limit=3

• Opt-in. No token, no join, no extra credit, no extra latency — a caller who never asks pays exactly what they always paid.
• Billed per row actually filled. The ceiling is held up front; a credit is kept only for a row a fresh sibling lookup filled. Rows served from the sibling’s cache are free, unfillable rows are refunded, and a page that joined in full is cached whole, so an immediate repeat is 0 credits.
• A row cap caps the bill. limit=3&include=profile on LinkedIn holds 22 credits, not 50.
• It tells you what it did. data.hydration reports rows, lookups, cache hits, credits held vs kept, and milliseconds; _warnings carries _partial when only some rows filled.

socialcrawl_endpoint with id: "hydration" lists every join, its price and the full contract; socialcrawl_estimate quotes one call with its include.

Monitors — schedule any recipe

socialcrawl_manage with area: "monitors" wraps any registry endpoint or Prism composite in a scheduled, stateful monitor (/v1/monitors/*). It re-runs the recipe hourly/daily/weekly (or on a cron), delivers each result to an HMAC-signed webhook, raises alerts on metric thresholds or changes, and keeps a per-run time-series you can read back. “Prism answers once; monitors watch it for you.” Managing monitors costs 0 credits; each scheduled run bills the recipe’s normal cost plus a 1-credit scheduling premium. See socialcrawl_endpoint with id: "monitors" for the full contract.

Web — scrape, crawl, browse

socialcrawl_manage with area: "web" drives the full web-scraping and browser-automation surface (the web platform, /v1/web/*) through a single action parameter:

• Sync reads — scrape (URL → markdown/HTML/screenshot/links), search (web search with page content), map (discover a site’s URLs), extract (LLM structured data from a page).
• Async jobs — crawl a whole site, batch_scrape many URLs, or agent (autonomous multi-step web task); each returns a job you poll with job_get/job_list, inspect with job_errors (per-page failures), and stop with job_cancel. crawl_preview dry-runs a crawl’s parameters for free before you pay for it.
• Monitors — monitor_create/list/get/update/delete/checks re-check a URL on a cadence and deliver changes to a webhook.
• Sessions — session_create/get/list, session_execute (run code in the live page), session_close.

Most of the paid web surface is metered rather than flat: a crawl holds limit credits and refunds every page it did not crawl; a session holds against ttl_seconds and settles on close. Job, monitor, and session management is 0 credits. See socialcrawl_endpoint with id: "web", or socialcrawl_estimate with id: "web".

Smart validation

Before making any API call, socialcrawl_request mirrors the backend’s own pre-billing validator against the bundled registry data: the platform and resource exist, required parameters and oneOf groups are satisfied, enum values are legal, integers sit inside their declared range, parameter couplings hold (order needs sort; Reddit’s timeframe needs sort=top), and CSV lists are within their entry limits. A malformed call fails instantly and for free instead of costing a round trip — and an agent gets told exactly what to fix rather than looping on a call that can never succeed.

Retry-safe requests

Pass an idempotencyKey to socialcrawl_request (UUIDv4 recommended) to make the call retry-safe. If the request is replayed within 24h, the server returns the original response and deducts 0 credits (X-Idempotent-Replay: true).

Supported Platforms

socialcrawl_find with no task lists every platform with its live endpoint count.

Error Handling

The MCP server handles errors gracefully and gives the agent actionable guidance. Every API error passes the server’s own error.message through, followed by reason: … when the API names one and request_id: req-… on its own line (read from the body, or the X-Request-Id header when the body is not JSON). Quote the request_id when you report a problem: it is how a call is found in the logs and the credit ledger.

Links

• Get Your API Key — 100 free credits, no credit card required
• API Documentation — full endpoint reference, credits, and error codes
• SocialCrawl Website
• npm Package
• MCP Registry
• Getting Started Guide
• How It Works