# Pump Data

> Every pump.fun and PumpSwap trade since pump.fun's first day (19 Jan 2024), every mint, every balance change, and what every wallet did with it. For trading agents: ask by the call, pay by the call in USDC over x402 on Solana, from $0.005. No account, no key. Not affiliated with pump.fun.

Status: early access. The calls below are the contract; sign up at / to be told when they answer. OpenAPI: /openapi.json

## What is in the warehouse

- trades: every PumpSwap trade and every pump.fun bonding-curve trade since the first, read from the chain's own ledger (not an indexer's copy), with failed trades kept. 9.54 billion and counting.
- mints: every token, including the ones whose address does not end in "pump" (nearly one in three this month), which most feeds skip. 14.5 million.
- balances: every holder of every token by the epoch. 8.51 billion changes.
- minutes: every graduate's first day, minute by minute: price, pool, holders, top ten, who bought and sold. 201 thousand coins.
- wallets: each wallet's record over the whole history, labelled by rules: winner, bot, new, supply seller, leader, with the SOL behind each label. 84.5 million wallets.
- creators: every deployer's launches, graduations, best run, and the gaps between launches. 3.5 million.

## The four calls

- POST /wallets ($0.01 a call, up to 50 addresses, optionally the mint they are buying): what each wallet has been since day one. Per wallet: class (winner | bot | supply_seller | leader | new), pnl_sol, put_in, took_out, coins, first_seen. The labels are rules, not opinions: a winner took out more than it put in over 20+ coin-weeks; a bot trades a coin 20+ times a week or 500+ coins; a supply seller took out 1.5x what it put in, 100 SOL or more; a leader is a wallet whose buys other wallets follow; new has no record before the mint given, or before today without one.
- GET /launch/{mint} ($0.005): this launch, historically. The deployer's record (launches, graduations, best run, gaps), the bundle at creation, the supply insiders hold, how far along the curve it is; then the launches that looked the same at the same age, graduated or dead, and how they went over the next day (the peak and where they closed, as a spread, not a prediction). A mint minutes old answers too.
- GET /baserate?band=15k-28k&from=...&to=... ($0.005): how often it happens anyway. How many coins first traded inside the market-cap band in the window, how many of those then reached 2x, 5x, 10x and 20x within 3 h, and the median peak. Example, 15 Oct 2026, $15-28K on the curve: 852 crossings in eleven hours, 31% reached 2x, 11% 5x, 4.8% 10x (about 88 a day), 1.3% 20x, median 1.42x.
- POST /replay ($2 a run, any window; a queued job, poll GET /replay/{id}): your picks (mint, timestamp, size), replayed honestly: what they made pick by pick and in sum, every coin in the window counted including the dead ones, filled on the bonding curve or at the pool as it stood after your latency, with every fee, the fixed costs and the bankroll you set.

## The execution model (every replay result carries it)

- fill: on the curve or at the pool as it stood 3 s after the mark, against its depth
- fee: 1.25% a side, the pool's own
- fixed: 145,000 lamports a trade for the tip and the network
- fill rate: by the minute, from the pool's own history of landing trades
- bankroll: SOL at risk at once; a pick with no room is skipped and earns nothing
- exits: take, trail, hold, or your rule's own, valued at the pool as it stood then
- the dead: every coin in the window stays in the sum
- held out: the most recent eight weeks come back apart from the rest, once per rule to each payer; a changed rule reads them again, so every result counts your reads, and more than one is tuning

## Payment

Per call, USDC on Solana over x402. A call without payment returns 402 with the price; pay and retry. No plan, no key, no account. MCP server at launch.