Who This Is For
This guide is for content creators, marketers, and founders who want real social media intelligence without paying for expensive analytics tools.
It is especially useful if you are:
Running a TikTok or Instagram account and want to know which posts actually drive reach vs. which just collect passive likes
Watching a competitor's content and trying to figure out what's working for them before making decisions about your own
Managing a brand across 2-4 accounts and wasting time manually scrolling to compare performance
Producing social reports for clients who want real data but won't pay for a dedicated analytics platform
Trying to find the hooks and hashtags that correlate with real reach in your niche, not just the ones that show up most often
Doing weekly content audits manually and looking for a faster, more repeatable way to read the signals
This is not for someone who posts casually and doesn't need pattern data to make decisions.
Apify + Live Artifacts makes the most sense when you are analyzing at least 10-20 posts at a time and plan to run the same analysis repeatedly across different accounts or time periods.
How the Workflow Runs
Connect Apify MCP → Scrape in chat → Collect JSON → Build artifact → Paste and analyze → Reuse the lens
Time to Set Up
Time to set up: 10-15 minutes for the first setup, Apify account, MCP connector, and a test scrape.
Daily time after that: 2-3 minutes per analysis run. Scrape in chat, paste the JSON into your artifact, read the results.
Tools Needed
Claude desktop app with Cowork mode enabled
Apify account (free tier includes $5 in platform credit every month, no credit card required)
Apify MCP connector (installed in Claude via Settings → Connectors)
A social media handle or hashtag to analyze (TikTok, Instagram, or YouTube)
Step-by-Step Walkthrough
Step 1: Understand what Apify + Live Artifacts should and should not do
Before setting anything up, get the architecture right. This is where most first attempts go wrong.
Apify + Live Artifacts can help with:
Scraping any public TikTok, Instagram, or YouTube account via ready-made scrapers (called Actors)
Generating a custom scoring and visualization dashboard tailored to your specific question
Reusing the same dashboard indefinitely: build the lens once, feed it new data anytime
Comparing multiple accounts on engagement rate, not raw follower counts, so a small account and a large one compare fairly
Finding which hooks, hashtags, or content formats pull real reach in a given niche
It cannot fully replace platform-native analytics for private accounts, paid-tier metrics, or real-time data.
It does not cover:
Auto-refreshing from Apify inside the artifact. The artifact cannot call Apify directly due to an authentication constraint (explained below). You scrape in chat and paste the JSON each time.
Private account data or any metrics behind a login
Platforms or account types without an Apify Actor available
Historical data beyond what the scraper can reach on public pages
Why the artifact can't call Apify itself. When you connect Apify, the authentication credential is held by Claude in your chat session. An artifact running in the browser makes a separate, unauthenticated request; Apify sees no credentials and returns nothing. The fix is simple: let Claude handle the scrape in chat, then paste the resulting JSON into the artifact. Same analysis, same visuals. The scrape just happens one step earlier, where the authentication actually lives.
Step 2: Set up Apify MCP in Claude
This takes about 5 minutes.
Create a free Apify account at apify.com. No credit card needed. You get $5 in platform credit every month that resets automatically, enough to run hundreds of small scrapes.
In Claude, go to Settings → Connectors → Add custom connector and enter
https://mcp.apify.com. Or search "Apify" in the connector directory and install it directly.Authorize via OAuth. A browser window opens, you sign into Apify and approve. No API token to paste.
Pin one Actor to start. In the connector's tool settings, enable only the scraper you'll actually use. For TikTok, start with
clockworks/tiktok-scraper.Test with a zero-cost call first. Open a new chat with Apify toggled on and ask: "Search the Apify Store for TikTok scrapers." This lists available Actors without running one, so it costs nothing. When it returns results, you're connected.
Note: The hosted connector at mcp.apify.com is the recommended path. It uses OAuth (no token to manage), updates itself automatically, and is easier to set up than a local install. Only use the local route if your Claude setup can't connect to remote MCP servers.
Important: Never paste your Apify API token into chat or hardcode it into an artifact. OAuth exists so you don't have to. A token typed into a conversation or embedded in artifact code is effectively public.
Step 3: Replace the default configuration with your real context
Before asking Claude to build anything, define what you're actually trying to learn.
Tell Claude your:
Social platform to analyze (TikTok, Instagram, YouTube; each needs a different Actor)
Account handle or hashtag to start with
The question you're trying to answer (e.g., "which of my posts get the most spread" or "how does my engagement rate compare to three competitors")
Metrics that matter to you (views, engagement rate, shares, comments, or a combination)
How many posts per run (start at 10, scale once the pipeline is working)
Whether you want to analyze one account deeply or compare multiple accounts side by side
This determines which type of dashboard to build in Step 4 and what fields the scraper needs to return.
Step 4: Teach Claude what a "good dashboard" looks like for you
→ Once your context is in, pick a dashboard type before asking Claude to build anything.
→ This matters because each type of analysis needs a different scoring formula and a different layout. A generic "build me a dashboard" prompt gives you a generic chart that doesn't answer a specific question well.
→ There are three proven starting points:
Viral Signal Spotter: Ranks your own posts by a weighted score where shares and comments count more than likes (because they signal active spread, not passive approval). Tags each post Hot, Rising, or Low. Shows your top post as a multiple of your account's median views. Best for analyzing a single account.
Competitor Benchmark Board: Pulls 2-4 accounts and normalizes everything to engagement rate so size doesn't distort the comparison. Shows cadence, average reach, and top posts side by side. Best for competitive analysis.
Hook and Hashtag Lab: Parses every caption and hashtag from a large batch, ranks hashtags by average views (not just frequency), and clusters opening-line patterns by the reach they pull. Best for content strategy.
→ Start with the Viral Signal Spotter. It's the simplest to build and test, and it's the foundation the other two extend.
→ The build prompts in the Prompt section below have all the technical guardrails already built in. Use them as-is.
Without this step, you may build a dashboard that looks right but silently breaks when you paste real data (mismatched field names, localStorage errors, a Tailwind class that doesn't compile in the artifact environment). With this step, the artifact is built correctly from the start.
The output of this step should ideally comprise:
A React artifact built in a Cowork session with a paste area for raw JSON
A scored, ranked table of posts with at least one chart
A dark/light toggle and an "explore with sample data" option so the dashboard works before your first real scrape
A Copy Scrape Prompt button you'll use for every future data pull
A forgiving JSON parser that handles field name variations across different Actors
Want the full breakdown?
This is where you get real AI workflows, prompts, and systems you can use to automate your work. If you're serious about using tools like Claude to grow your business, this is for you.
Unlock full access today