A big account tweets, the trenches launch it fifteen times. Bouncer lets the OG in.
Paste a ticker, a name or the X post. Bouncer lists every pump.fun coin on it in launch order, flags the copies, the bundles and the dev dumps, and points at the one to ape.
Tickers launched more than once on pump.fun in the last hours. The first one leads, the copies queue behind it.
The hour's most traded pump.fun coins, read from the chain. Refreshed every 90 seconds.
Three things your agent gets, and none of them asks you to trust the agent.
deny_if and review_if rules on 23 metrics, in plain JSON. Omit it and the default policy runs.
Write a policyALLOW, REVIEW or DENY. A metric it can't read never passes: its rule goes to REVIEW.
Gate a coinEvery check, every number, the policy hash and the slot, signed ed25519. Verify it anywhere.
Verify a receiptSend your own rules to change any of it. These run when you send none.
The creator can still print more supply.
mint_authority == trueThe creator can still freeze your wallet.
freeze_authority == trueThe dev still holds more than a fifth.
dev_holds_pct > 20Bundled wallets bought more than a quarter at launch.
bundle_pct > 25Liquidity under $5k, top 10 over 40%, snipers over 10%, a dev who sold, under 20 swaps an hour.
review_ifThe same call your agent makes, with the default policy. The first answer is a real pump.fun coin read when this page loaded.
25 free calls a day per IP. After that the API asks for 0.01 USDC over x402.
Before a buy, the agent calls gate_check with the mint and, if you want, your policy.
Bouncer reads the token from Solana and Jupiter: authorities, dev wallet, bundles, snipers, holders, liquidity, flow.
Every rule is checked in code. One deny rule hit is a DENY. A review rule hit, or a metric it can't read, is a REVIEW.
The receipt carries the checks, the metrics, the policy hash and the slot, signed with the key at /api/v1/key.
Read from Bouncer's own log of signed receipts when the page loads.
Every gate an agent asked for
DENY verdicts: buys that never happened
Agents asking before they buy
REST for any agent framework. The receipt verifies with the public key, offline or through /api/v1/verify.
| Endpoint | What it does |
|---|---|
POST /api/v1/gate | {token, policy?} → verdict and signed receipt. 25 free a day, then x402. |
POST /api/v1/verify | {receipt, signature} → valid or not. |
POST /api/v1/watch | {token, policy?, webhook_url} → a new signed receipt to your webhook when the verdict changes, for 72 h. |
GET /api/v1/metrics | The 23 metrics and the default policy. |
GET /api/v1/key | The ed25519 public key that signs every receipt. |
Remote MCP over streamable HTTP. Add it to Claude Code, Claude Desktop, Cursor or any MCP client.
claude mcp add --transport http bouncer {host}/mcp| Tool | What it does |
|---|---|
gate_check | A token and a policy in, ALLOW / REVIEW / DENY and a signed receipt out. |
list_metrics | Every metric a rule can use, and the default policy. |
verify_receipt | Checks a receipt's signature. |
watch_token | Re-gates a token every 5 minutes and posts to a webhook when the verdict changes. |
Launched on pump.fun. The contract address is published here and on X at launch, nowhere else first.
Drops at launchNo. It only answers. Your agent keeps its own wallet and decides what to do with the verdict.
That rule goes to REVIEW, never to ALLOW. The receipt says read_complete: false.
Canonical JSON of the receipt, ed25519 signature, public key at /api/v1/key. Any ed25519 library verifies it, or POST it to /api/v1/verify.
25 calls a day free. After that the API answers 402 with x402 terms: 0.01 USDC per call, paid on Solana, checked on chain before the answer.
No. It means the coin passed your rules at that slot. Rules describe a launch, they don't predict a price.
One MCP line, your rules, and a signed receipt for every coin your agent wanted.
