How to research creators with Claude (5 prompts that return receipts)

Full disclosure: we make Eden, and these prompts run on its index. The prompts themselves describe the analysis any good researcher would do; the MCP is what makes them take seconds instead of an afternoon.

Why "just ask Claude" fails without data

Ask a plain Claude chat "what's working on LinkedIn right now" and you get confident, generic advice: post consistently, use hooks, tell stories. It is not lying, it is guessing, because it cannot see a single current post.

The fix is giving it an index. Eden's MCP connects Claude to millions of indexed posts across X, LinkedIn, Instagram, TikTok, YouTube, and Substack, with every post scored against its own creator's normal engagement. That last part matters: 5,000 likes is routine for a huge account and a breakout for a small one. The outlier score ("this did 8.4x their baseline") is what separates research from scrolling.

Setup takes about a minute: add https://mcp.eden.so/mcp as a connector in Claude and sign in, or from the terminal:

claude mcp add --transport http eden https://mcp.eden.so/mcp

The quick start has the click-by-click version. Now the prompts.

Prompt 1: analyze a creator against their own baseline

Analyze @creator on X. What are their top outliers from the last 90 days, and what do those posts have in common?

What comes back: the account's typical engagement, the posts that beat it by the biggest multiple, and the pattern across them (hook shape, format, topic). Because the scoring is against the creator's own normal, you learn what worked for them, not what a big follower count made look good.

If Eden has never seen the account before, it starts indexing in the background and tells you so. Ask again in a few minutes and the full analysis is there.

Prompt 2: sweep a niche for breakout posts

Find posts about home fitness that broke out in the last 30 days across X and Instagram. Only show me outliers, and include the multiplier for each.

What comes back: real posts with receipts. Text, author, platform, and the outlier score, so "3x their baseline" small-account breakouts surface next to the big names. This is the prompt that replaces an hour of scrolling with a ranked shortlist of proven ideas, and it is the core of what a social research MCP does.

Prompt 3: roll up a creator list

Look at my "Productivity writers" list. What did each creator post this week, and whose posts overperformed?

What comes back: a digest across every creator in the list, with the overperformers flagged. If you track competitors or inspirations in Eden creator lists, this turns the whole list into one weekly briefing. You can also attach the list at the start of a chat so every follow-up question stays scoped to that group.

Prompt 4: compare two creators head to head

Compare @creatorA and @creatorB on LinkedIn. Who gets more outliers relative to their size, and how do their hooks differ in the last 30 posts?

What comes back: a side-by-side on baseline, outlier rate, cadence, and hook style. Raw follower counts make this comparison meaningless; baselines make it fair. This is the prompt to run before you pick which competitor's playbook deserves your study time.

Prompt 5: turn the findings into drafts

Take the three strongest patterns from this research and write three posts for my topic, which is career changes after 40. Save them to my Ideas board as drafts.

What comes back: drafts modeled on shapes that already worked, saved into your Eden workspace (the connection's write access is what allows the save; you approve it once when connecting). From there you can also say "schedule the best one for Thursday morning" and close the loop in the same chat. That handoff from research to queue is covered in scheduling posts from Claude.

Why not just use a scraper MCP?

Apify, Bright Data, and similar tools also connect data to agents, and for developers building datasets they are the right infrastructure. But they return raw, unscored data: posts and metrics as JSON, at scraper pricing, often with per-platform setup and API keys.

Raw data leaves the hard part undone. Nothing tells the agent which posts overperformed, against whose baseline, or by how much, so "big account posted it" and "this actually broke out" look identical. A social research MCP hands the agent decisions, not exports: the corpus is already indexed, already scored, and attached to scheduling so the finding can become a post without leaving the conversation. Our MCP comparison covers the scraper category in more depth.

Frequently asked questions

What is a creator research MCP?

An MCP server that gives an AI assistant real creator data it can query: indexed posts, engagement scored against each creator's own baseline, and profile-level analysis like top categories and cadence. It is the research half of what we call a social research MCP; the other half is publishing what the research finds.

Do I need API keys or a scraper setup?

No. Eden's MCP is hosted: you connect one URL and sign in through the browser. There are no platform API keys, no proxies, and nothing to host.

What if the creator I ask about is not in the index?

Eden starts collecting their content in the background the first time anyone asks. The first answer may be thin; ask again in a few minutes and the analysis fills in. Accounts with thousands of posts take longer to backfill, and the response includes a time estimate.

Which platforms does this cover?

X, LinkedIn, Instagram, TikTok, YouTube, and Substack. The same prompts work across all of them, so you can also ask cross-platform questions like "did this creator's idea work better on X or LinkedIn?"

Can I do this on a free plan?

Yes. The Eden MCP connects on the Free plan ($0, no card), and Claude's custom connectors are where you add it. Heavier research runs into free-tier limits eventually; paid plans start at $29/mo.

Try one prompt now. Start free, connect the Eden MCP, and run prompt 1 on the creator you learn the most from. The pattern in their outliers is usually visible in the first reply.