Social Media MCP Servers: Social Data for AI Agents, Compared
How a social media MCP server gives Claude, ChatGPT or Cursor live data from Reddit, X, TikTok, Instagram and YouTube, and how FetchLayer, ScrapeCreators, SociaVault, Apify, Bright Data and ScraperAPI compare.
Short answer: a social media MCP server gives an AI agent tools such as “search Reddit” or “get this TikTok account’s videos”, so Claude, ChatGPT, Cursor or your own agent can fetch live public data mid-conversation. FetchLayer’s MCP server at https://mcp.fetchlayer.dev exposes 244 tools across 37 sources: Reddit, X, YouTube, TikTok, Instagram, Weibo and Douyin, plus app stores, reviews, ad libraries, marketplaces and search engines. It bills like our REST API: one credit per page of results, $1.99 per 1,000 pay-as-you-go, with 30 free requests. Other providers’ servers, checked on 14 November 2026, are compared below.
What a social media MCP server does
The Model Context Protocol lets an AI client discover tools on a server and call them. For social data, that means the agent can:
- Look things up itself. Ask “what are people on Reddit and TikTok saying about standing desks this week?” and the agent picks the search tools, reads the results and answers with sources.
- Chain calls. Find a creator, read their latest videos, then pull the comments on the most-watched one, without you writing glue code.
- Work inside the tools you already use. The same server works in Claude, ChatGPT, Claude Code, Cursor, VS Code, Windsurf and your own code.
The REST API is the better fit for scheduled jobs and pipelines. MCP is for when a model decides what to fetch next.
What FetchLayer’s MCP server covers
Every endpoint is a tool, named for its platform and action, such as tiktok_user_videos, instagram_post_comments or scrape_search for Reddit. Tool counts by source, from the live MCP docs:
- Social and video: Reddit (17), X/Twitter (13), Instagram (11), Douyin (10), Weibo (9), TikTok (8), YouTube (5), Link in Bio (5)
- App stores and reviews: App Store (2), Google Play (2), Trustpilot (5), Glassdoor (6), Tripadvisor (5), Goodreads (12), Google Maps (4)
- Ad libraries: Facebook (7), Google (7), TikTok (7), LinkedIn (6)
- Shopping: Amazon (4), eBay (6), AliExpress (6), Temu (3), Shopify (9), WooCommerce (9)
- Search and AI: DuckDuckGo (8), Bing (6), AI mentions in ChatGPT, Gemini, Google AI Overviews, Muse / Meta AI and more (6)
- Communities, work and travel: Product Hunt (8), Printables (11), MakerWorld (9), Upwork (3), Email Finder (2), Airbnb (5), Booking.com (4), Flight Search (1), Web Unblocker (2)
One more tool, resolve_url_type, is listed whichever sources you choose: it turns a Reddit link into the tool and input that read it.
That breadth is the point of one server: an agent researching a product can read Reddit threads, TikTok videos, App Store reviews and the brand’s ads in one session, with one key and one bill.
Too many tools? Load only what you need
Some clients cap how many tools they accept. Windsurf loads at most 100 across all servers, VS Code and the OpenAI API at most 128 per request, and ChatGPT picks tools less reliably past about 70. Add ?platforms= to the address and the connection only sees those sources:
https://mcp.fetchlayer.dev/mcp?platforms=reddit,tiktok,instagramThat address lists 37 tools instead of 244.
Connect it
- Claude (web, desktop, mobile), ChatGPT, Claude Code, Codex, Gemini CLI, VS Code and Cursor support MCP sign-in. Add
https://mcp.fetchlayer.devas a server with no key, and the client opens a FetchLayer page where you approve it. - Clients configured with a file take your API key as a header. In Claude Code, for example:
claude mcp add --transport http fetchlayer "https://mcp.fetchlayer.dev/?platforms=reddit,tiktok,instagram" \
--header "Authorization: Bearer ss-your-fetchlayer-key"Step-by-step setup for Claude, Claude Code, Cursor, Windsurf, VS Code and others is in the MCP setup hub. It shows Reddit, but the steps are the same for any platform.
Call it from your own agent code
The official MCP Python SDK connects your own agent the same way a desktop client does. This script loads only the Reddit, TikTok and Instagram tools and runs one search on each:
import asyncio
import json
import os
from mcp import ClientSession
from mcp.client.streamable_http import create_mcp_http_client, streamable_http_client
# Only the platforms this agent needs, so the client loads a short tool list
URL = "https://mcp.fetchlayer.dev/mcp?platforms=reddit,tiktok,instagram"
HEADERS = {"Authorization": f"Bearer {os.environ['FETCHLAYER_API_KEY']}"}
TOPIC = "standing desk"
async def main():
async with create_mcp_http_client(headers=HEADERS) as http:
async with streamable_http_client(URL, http_client=http) as (read, write, *_):
async with ClientSession(read, write) as session:
await session.initialize()
tools = (await session.list_tools()).tools
print(len(tools), "tools for reddit, tiktok and instagram")
calls = {
"scrape_search": {"query": TOPIC, "sort": "new"},
"tiktok_search_videos": {"query": TOPIC},
"instagram_reels_search": {"keyword": TOPIC},
}
for name, args in calls.items():
result = await session.call_tool(name, {**args, "format": "json"})
data = json.loads(result.content[0].text)
items = data.get("items") or data.get("videos") or data.get("posts") or []
print(f"{name:<24} {len(items):>3} results")
asyncio.run(main())Our run on 14 November 2026 (pip install mcp, version 2.3.0):
37 tools for reddit, tiktok and instagram
scrape_search 7 results
tiktok_search_videos 13 results
instagram_reels_search 12 resultsIn a real agent, you’d hand tools to your model and let it choose the calls. The script calls them directly to show the shape of each result.
Prompts that use several platforms
Search Reddit and TikTok for “standing desk” from the last month. Group the complaints by theme, show how many posts or videos mention each, and link the strongest examples.
Compare @nasa’s latest 20 Instagram posts with its latest 20 TikTok videos: posting frequency, average likes, and which topics do best on each.
Find the five most-discussed AI note-taking apps on Reddit this month, then pull each one’s recent one-star App Store reviews and summarize what they share.
Social media MCP servers compared
Read on each provider’s own pages and docs on 14 November 2026:
| Provider | Hosted address | Sign-in | Social coverage | Tools | Billing |
|---|---|---|---|---|---|
| FetchLayer | mcp.fetchlayer.dev | OAuth, or API key header | Reddit, X, YouTube, TikTok, Instagram, Weibo and Douyin, among 37 sources | 244, one per endpoint | Same as the API: 1 credit per page; 30 free requests |
| ScrapeCreators | api.scrapecreators.com/mcp | OAuth, or x-api-key | “27+ social media platforms”, including LinkedIn, Facebook, Threads, Bluesky and Pinterest | “110+” endpoints as tools | API credits, “1 credit === 1 request (for most endpoints)”; 100 free credits |
| SociaVault | mcp.sociavault.net/mcp, or npx | API key in the URL or a header | “11 platforms”, including LinkedIn, Facebook, Threads and Pinterest | 107 | API credits, most endpoints 1 per request; 50 free |
| Apify | mcp.apify.com, or npx | OAuth, or API token | Any Store actor, such as its Instagram and TikTok scrapers | 49, mostly generic (search-actors, call-actor) | Each actor’s own per-result price; $5 free a month |
| Bright Data | mcp.brightdata.com/mcp, or npx | Token in the URL, or OAuth | LinkedIn, Instagram, TikTok, Facebook, YouTube, X, Reddit, through generic scraper tools | 30 | $1.50 per 1,000 results; 5,000 free requests a month |
| ScraperAPI | mcp.scraperapi.com/mcp, or pip | API key header | None: Google, Amazon, Walmart, eBay and Redfin | 28 | ScraperAPI credits |
How to choose:
- You need LinkedIn, Facebook, Threads or Pinterest: ScrapeCreators or SociaVault cover social networks we don’t. Our ScrapeCreators comparison goes platform by platform.
- You want one typed tool per action: FetchLayer, ScrapeCreators and SociaVault give the model a named tool for each endpoint. Apify and Bright Data route most social data through generic “find a scraper, then run it” tools, which take more steps and depend on the model choosing the right scraper.
- You need more than social: FetchLayer adds app stores, review sites, ad libraries, marketplaces and search engines on the same server and bill.
- You need the cheapest pages of posts or comments: see our TikTok and Instagram price comparisons. One FetchLayer request returns a whole page of videos or comments.
What it costs
MCP calls bill exactly like the REST API: one credit per page of results, $1.99 per 1,000 pay-as-you-go, or $0.50–$1.67 per 1,000 on monthly plans. The script above made 3 calls. A research session where an agent runs 20–40 searches and reads a few threads costs well under $0.10, and failed calls aren’t billed.
FAQ
Should I use MCP or the REST API?
Both reach the same endpoints at the same price. Use MCP when a model decides what to fetch next, and the REST API for scheduled jobs and pipelines where your code decides.
Does it work with ChatGPT?
Yes. Add https://mcp.fetchlayer.dev as a connector and sign in. ChatGPT picks tools less reliably past about 70, so add ?platforms= with the sources you need.
Do I need a separate key per platform?
No. One FetchLayer key covers every source, and you can restrict a key to some platforms in the dashboard.
Can the agent post or send messages?
No. Every FetchLayer tool reads public data. Nothing posts, likes or messages.