Most pumpfun APIs return new coins. A different question is more interesting to a developer: which accounts made the most realized profit over the last 30 days, and which coins are several tracked traders buying right now?
This post answers both with two plain GET requests and about 100 lines of TypeScript. No key, no scraper, no library. The output is two small tables, and a saved fixture lets you reproduce the numbers printed below.
Disclosure: I am the founder of copy.top. The API used here is copy.top's own free public API. copy.top is independent and not affiliated with pump.fun or fomo.
Setup
The files are TypeScript and run on Node 22.6 or newer with type stripping (tested here on Node 24.7):
node --experimental-strip-types run.ts Imports between files need the .ts extension, and types use import type, which this mode requires. For plain JavaScript, rename to .mjs, delete the annotations and the import type lines.
The two endpoints
The REST API is documented on the developers page and, for pumpfun specifically, on the pumpfun API page. Everything is a GET that returns JSON, with no key. The public limit is about 20 calls a minute and answers are cached for up to a minute, so there is no point in polling faster than that.
-
GET https://copy.top/v1/traders?platform=pumpfun&window=30d&sort=realized&limit=10returns a leaderboard.windowis the period,sort=realizedranks by realized profit,limitis the number of rows. -
GET https://copy.top/v1/coins/hot?window=1hreturns coins that tracked traders bought in the last hour, with counts and dollar totals.
I checked the field names below against real responses. A trader row has handle, name, pnl, realized, unrealized, roi, winRate (a number from 0 to 1), closed (closed positions), followers, copiers and a few more. A coin row has symbol, token (the address), traders, topTraders, topStillIn, sold, boughtUsd, soldUsd, firstBuyMinAgo, lastBuyMinAgo and buyers. A handle that starts with pf_ is a pumpfun account.
Types
Only the fields the script reads.
// types.ts export interface Trader { handle: string; name: string; pnl: number; realized: number; unrealized: number; roi: number; winRate: number; // 0 to 1 closed: number; // closed positions in the window followers: number; copiers: number; url: string; } export interface Leaderboard { window: string; sort: string; traders: Trader[]; } export interface HotCoin { chain: string; token: string; // address symbol: string; mcap: number | null; traders: number; // tracked traders who bought it in the window topTraders: number; // how many of them are top traders topStillIn: number; // top traders still holding sold: number; boughtUsd: number; soldUsd: number; firstBuyMinAgo: number; lastBuyMinAgo: number; buyers: string[]; } export interface HotCoins { window: string; at: string; coins: HotCoin[]; } Step 1: a small fetch wrapper
The wrapper does three things: it builds the query string, it stops a hung request after 20 seconds with AbortSignal.timeout, and it throws on a non-200 answer. The second function waits 1.5 seconds before it calls, which keeps a run far below the public limit.
// api.ts import type { Leaderboard, HotCoins } from "./types.ts"; const BASE = "https://copy.top/v1"; const DELAY_MS = 1500; // polite: the public limit is about 20 calls a minute const sleep = (ms: number) => new Promise((r) => setTimeout(r, ms)); async function get<T>(path: string, params: Record<string, string | number>): Promise<T> { const qs = new URLSearchParams(Object.entries(params).map(([k, v]) => [k, String(v)])); const res = await fetch(`${BASE}${path}?${qs}`, { headers: { accept: "application/json" }, signal: AbortSignal.timeout(20_000), }); if (!res.ok) throw new Error(`${path} answered ${res.status}`); return (await res.json()) as T; } export async function getLeaderboard(limit = 10): Promise<Leaderboard> { return get<Leaderboard>("/traders", { platform: "pumpfun", window: "30d", sort: "realized", limit, }); } export async function getHotCoins(): Promise<HotCoins> { await sleep(DELAY_MS); return get<HotCoins>("/coins/hot", { window: "1h" }); } A run makes exactly two requests.
Step 2: the runner
With no argument the script calls the live API. With --fixture it reads fixture.json instead and makes no network call. With --save it also writes the live answers to fixture.json.
The second table filters the coins with traders >= 3, so only coins where several tracked traders bought in the last hour remain. Some coins have a blank or invisible symbol, so label falls back to the first characters of the address.
// run.ts import { readFileSync, writeFileSync } from "node:fs"; import { getLeaderboard, getHotCoins } from "./api.ts"; import type { Leaderboard, HotCoins } from "./types.ts"; const useFixture = process.argv.includes("--fixture"); const save = process.argv.includes("--save"); let board: Leaderboard; let hot: HotCoins; if (useFixture) { ({ board, hot } = JSON.parse(readFileSync("fixture.json", "utf8"))); } else { board = await getLeaderboard(10); hot = await getHotCoins(); if (save) writeFileSync("fixture.json", JSON.stringify({ board, hot }, null, 2)); } const usd = (n: number) => Math.round(n).toLocaleString("en-US"); // some symbols are blank or invisible characters: fall back to the start of the address const label = (c: { symbol: string; token: string }) => /[A-Za-z0-9]/.test(c.symbol) ? c.symbol : c.token.slice(0, 6); console.log(`top ${board.traders.length} pumpfun traders, ${board.window}, sorted by ${board.sort}`); console.table( board.traders.map((t, i) => ({ rank: i + 1, handle: t.handle, name: t.name, realized_usd: usd(t.realized), win_rate: `${Math.round(t.winRate * 100)}%`, closed: t.closed, })), ); // several traders: at least 3 tracked traders bought the coin in the last hour const several = hot.coins.filter((c) => c.traders >= 3).slice(0, 8); console.log(`\ncoins bought by 3 or more tracked traders, last ${hot.window} (as of ${hot.at})`); console.table( several.map((c) => ({ coin: label(c), traders: c.traders, top: c.topTraders, top_still_in: c.topStillIn, bought_usd: usd(c.boughtUsd), sold_usd: usd(c.soldUsd), last_buy_min_ago: c.lastBuyMinAgo, })), ); Live use is node --experimental-strip-types run.ts.
Reproduce the numbers
Live data changes every minute, so the output of a live run will not match this article for long. I ran the script once against the live API with --save, which wrote fixture.json, and then printed the output below from that fixture:
node --experimental-strip-types run.ts --fixture If you copy fixture.json next to run.ts (it is the raw JSON of that one live run, 2026-11-19), you get the same two tables, line for line.
Output
Exactly as printed:
top 10 pumpfun traders, 30d, sorted by realized ┌─────────┬──────┬───────────────────┬───────────────────┬──────────────┬──────────┬────────┐ │ (index) │ rank │ handle │ name │ realized_usd │ win_rate │ closed │ ├─────────┼──────┼───────────────────┼───────────────────┼──────────────┼──────────┼────────┤ │ 0 │ 1 │ 'pf_c311d04abf73' │ 'supermandev' │ '794,629' │ '97%' │ 34 │ │ 1 │ 2 │ 'pf_5ae8d0344ba0' │ 'NateNateNateNat' │ '395,711' │ '95%' │ 1002 │ │ 2 │ 3 │ 'pf_39b00d567226' │ 'n91' │ '276,107' │ '96%' │ 1020 │ │ 3 │ 4 │ 'pf_79183226f2be' │ 'alphaprophet' │ '273,777' │ '97%' │ 1022 │ │ 4 │ 5 │ 'pf_2ccf3a09cd0e' │ 'TarShortRiver' │ '255,298' │ '55%' │ 58 │ │ 5 │ 6 │ 'pikachu34' │ 'pikachu34' │ '241,712' │ '50%' │ 22 │ │ 6 │ 7 │ 'pf_fe83ee057073' │ 'MintGarnetGamer' │ '207,846' │ '55%' │ 56 │ │ 7 │ 8 │ 'pf_36908e8f0385' │ 'moan7' │ '190,064' │ '95%' │ 827 │ │ 8 │ 9 │ 'pf_921ca6d4fac4' │ 'UpbeatAloudChef' │ '165,123' │ '56%' │ 50 │ │ 9 │ 10 │ 'pf_5482d0f6ab22' │ 'MemeHubCo' │ '114,098' │ '32%' │ 41 │ └─────────┴──────┴───────────────────┴───────────────────┴──────────────┴──────────┴────────┘ coins bought by 3 or more tracked traders, last 1h (as of 2026-11-19T09:55:16.608Z) ┌─────────┬──────────────┬─────────┬─────┬──────────────┬────────────┬───────────┬──────────────────┐ │ (index) │ coin │ traders │ top │ top_still_in │ bought_usd │ sold_usd │ last_buy_min_ago │ ├─────────┼──────────────┼─────────┼─────┼──────────────┼────────────┼───────────┼──────────────────┤ │ 0 │ 'DCTY' │ 55 │ 15 │ 11 │ '265,881' │ '112,627' │ 1 │ │ 1 │ 'APOCRYCOIN' │ 27 │ 6 │ 4 │ '11,814' │ '9,306' │ 2 │ │ 2 │ 'DacityCoin' │ 19 │ 7 │ 4 │ '19,820' │ '5,118' │ 4 │ │ 3 │ '7Aasrt' │ 9 │ 4 │ 3 │ '2,472' │ '991' │ 1 │ │ 4 │ '3B8dDC' │ 33 │ 12 │ 3 │ '22,079' │ '13,040' │ 3 │ │ 5 │ 'GRS' │ 10 │ 6 │ 3 │ '2,533' │ '1,312' │ 8 │ │ 6 │ '48r1TbEV' │ 8 │ 3 │ 2 │ '1,545' │ '1,435' │ 2 │ │ 7 │ 'Sparrow' │ 15 │ 3 │ 2 │ '3,450' │ '1,760' │ 1 │ └─────────┴──────────────┴─────────┴─────┴──────────────┴────────────┴───────────┴──────────────────┘ Reading the tables
In the first table, check one thing by hand: realized_usd is the profit from positions closed in the 30 day window, so it is a sum of finished trades, not a value of what the account holds. The raw pnl field also includes open positions, which are marked at the current price. That is why the sort is on realized and not on pnl.
Look at the win_rate and closed columns together. Row 1 has a 97% win rate over 34 closed positions, and row 4 has 97% over 1022. Those are very different amounts of evidence, and a win rate says nothing about how large the losses are compared with the wins. Rows 5, 6, 7, 9 and 10 show win rates between 32% and 56%, next to large realized totals: a low win rate can still make money if the wins are big.
In the second table, traders is how many tracked traders bought the coin in the last hour, top is how many of those are top traders, and top_still_in is how many of them still hold. Compare bought_usd with sold_usd: for every coin on the list, both are well above zero, so people are already selling while others buy. The last column says how recent the latest buy is. These are counts of past activity at one moment, not a list of things to buy.
Where to take it next
Add GET /v1/traders/{handle} to read one trader in detail, or poll GET /v1/alerts with the cursor it returns. Writing the results to a file once a day lets you compare rankings from one week to the next.
The same data is exposed as an MCP server at https://copy.top/mcp, if you would rather query it from an AI agent than from code.
What this does not tell you
- Realized profit is a window, not a record. A 30 day number is a snapshot of one month. The person at the top this month can be absent next month.
- A win rate is not an edge. A high rate can come from many tiny wins and a few big losses. Always read it next to the number of closed positions.
- Buying is not endorsing. A tracked trader buying a coin tells you that one account bought it, at a price you do not get, at a time that has passed. Others in the same hour were already selling.
- Cached and delayed. Answers can be up to a minute old, and the trades behind them are older still.
Past performance does not predict results. Nothing here is financial advice, and copying a trader can lose money, including everything you put in.
diogenicious