KOL wallet tracking with Claude: 7 prompt recipes

KOL wallet tracking on Solana is one of the highest-density alpha sources in the market. The serious memecoin traders post their wallets publicly (as flex), and tailing the top ~30 of them has been a known-positive-EV strategy for two years. The constraint has always been the same: how do you ingest 1,400+ wallets' activity into something you can actually scan in real time?

Pre-MCP, the answer was a $99/mo Cielo subscription, a custom Helius webhook receiver, or a half-built Discord bot wired into your own indexer. Post-MCP, the answer is one prompt in Claude. This post is seven recipes you can paste into a Claude Desktop conversation (with techkern MCP installed) to get a working tracker by the end of the page.

The MCP tool you need

After installing techkern MCP, the single most relevant tool is:

• query_kol_wallets(window_minutes?, min_size_usd?, wallet?) — recent tracked-KOL buys

Backed by 1,400+ curated wallet addresses, every entry tagged with handle, size, mint, and live 30-day PnL. Query p95: 38ms. 8.4M+ queries served today.

Supporting tools you will compose with:

• mint_authority_check, holder_distribution, dev_token_history, query_solana_swaps

Recipe 1 — Fresh KOL entries on fresh launches

> "Every 5 minutes, pull KOL wallet buys from the last 10 minutes with size > $1,000. For each buy, if the mint is less than 30 minutes old, return the KOL handle, mint, size, and how long ago the mint launched."

This is the bread-and-butter. Fresh launches with KOL entries are the highest-frequency alpha shape on the platform.

Recipe 2 — KOL cluster alerts

> "Every 10 minutes, pull KOL buys from the last 30 minutes. Group by mint. If three or more distinct KOL wallets bought the same mint within the window, alert me with the mint, the list of KOLs, and total combined size."

Cluster signals are higher conviction than single-KOL alerts. Three serious traders converging on the same mint inside a 30-minute window is almost always either coordinated alpha or a coordinated exit-liquidity setup — both worth knowing about immediately.

Recipe 3 — KOL "rotates" detector

> "Every 30 minutes, pull KOL activity from the last 2 hours. For each KOL wallet, list any mint where they bought and also sold within the window. Output: wallet, mint, time-to-flip, realized P&L."

KOL rotates (entry → exit inside 2 hours) are the highest-confidence "this was always a scalp" signal. Useful for two reasons: (a) you avoid bag-holding after a KOL has clearly already exited, (b) the rotates themselves often signal that a token is dying right now.

Recipe 4 — KOL accumulation flag

> "Every hour, pull KOL buys from the last 24 hours. Find any mint where a single KOL bought 3+ times in the window with no sells. Output: wallet, mint, total accumulated size, number of distinct buy transactions, average entry price."

The opposite of recipe 3. A KOL averaging into a mint over multiple sessions is a much stronger signal than a single shot. Usually means they have conviction in something most of the market is sleeping on.

Recipe 5 — Mirror with cooldown

> "Watch wallet Ansem...Lk2 specifically. Every 2 minutes, check if they have opened a position in the last 5 minutes with size > $5,000. If yes, run mint_authority_check + holder_distribution on the mint and return a one-paragraph 'enter or skip' verdict — only flag enter if mint authority is revoked, dev hold is between 5%-50%, and top-10 hold is under 40%."

The classic mirror-trade prompt. The model adds a sanity-check layer (auth revoked + healthy holder distribution) so you do not blindly tail into something with obvious rug signals even if the KOL got in.

Recipe 6 — KOL leaderboard refresh

> "Pull KOL wallet activity from the last 30 days. For each wallet in my tracked list, compute realized PnL by summing buy and sell sizes against the current mid price for any open positions. Rank descending. Output the top 20."

Run this weekly. The KOL leaderboard shifts faster than you would expect — wallets that printed in January might be net-down by April, and rookies emerge. This recipe gives you a fresh leaderboard without any custom tracking infra.

Recipe 7 — KOL × dev-wallet cross-reference

> "For any KOL buy in the last 24 hours where the mint is less than 6 hours old: run dev_token_history on the mint's creator wallet. Flag any case where the dev wallet has at least one prior launch that graduated to Raydium AND zero prior launches that confirmed rugs. Output: KOL handle, mint, dev wallet, prior graduation count, prior rug count, dev's success rate."

The compound signal. Fresh launches that survived AND have KOL backing AND have a dev with track record are the rarest setup on the platform. When all three line up, the EV is unusually clean. This recipe surfaces them automatically every time you run it.

How to wire these into actual alerts

The seven recipes above all produce structured output you can parse. To turn them into push alerts, two paths:

1. Polling from a Node script. Use the MCP SSE transport directly via the OpenAI Agents SDK (TypeScript example in our install guide). Send any non-empty result to a Telegram bot, Discord webhook, or your phone via Pushover.
2. Polling from inside Claude. Run Claude in a long-lived terminal session, paste the recipe as a system prompt, and let it call query_kol_wallets on a self-scheduled loop. Pipe its alerts to your messaging surface.

We use approach 1 internally. Approach 2 is easier to set up for one-off experiments.

What this replaces

• Cielo Finance subscription: $99/mo
• Custom Helius webhook receiver: ~2-3 days of setup + ongoing maintenance
• Discord bot wired into your own indexer: never quite finished, breaks every other week

With techkern MCP: $19/mo Pro tier (or pay-per-query in $TECH), zero infra, every recipe above in under an hour.

The other thing this replaces is the thinking you had to do to design these recipes in the first place. Once you have structured access to KOL data via MCP, the prompts above are not the limit — they are the floor. The next 30 recipes are five minutes of conversation away.