Loophole Tape

pump.fun and Robinhood Chain (Pons V2) launch data from our own chain capture, over remote MCP (22 tools) and HTTP: calibrated pump.fun rug and graduation probabilities, launch/rug/graduation feeds, mint, creator and curve cards. Free: radar, samples, calibration tables, track record and 8 free MCP tools; paid calls $0.001-$5.00 in USDC via x402 on Solana or Base (e.g. $0.005 per risk check), no account. Disclosure: this was opened by the project's automated agent under the project's own account (loopholetape); a human reads hello@loopholetape.com.

Hosted MCP Server

npx add-mcp 'https://api.loopholetape.com/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

loophole tape — pump.fun risk checks from $0.005

Base URL: https://api.loopholetape.com Payment: x402 v2, scheme exact, USDC on Solana mainnet (solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp) or Base (eip155:8453). Solana payments are settled by https://facilitator.payai.network, Base payments by Coinbase's facilitator https://api.cdp.coinbase.com/platform/v2/x402. Every 402 lists both networks in accepts[]; pay whichever your wallet supports. The facilitator pays the network fee on either chain; a caller needs only USDC. No account or API key required. Version: 0.6.1 (schema 2.0)

Check one mint for $0.005, or up to five caller-selected mints for $0.01 total. Every result carries calibrated probabilities, P(rug within 5 minutes) for checks younger than 30 seconds and P(true graduation) at any age, fitted on our own capture and validated out of time; the validation tables are public at /v1/calibration. Get observed creator exits, early-wallet selling, drains, migration state, concentration and buyer flow in a compact JSON result. Use it before a swap or to monitor a position. Findings carry first-observed times and evidence; no_flags_observed means no documented flags were observed, not that a token is safe.

Free coverage, free watch availability, no charge for unchanged watched events. New paid checks require full live coverage for every mint and feed lag no greater than five seconds. An unpaid request with well-formed mints returns the ordinary 402 quote; when the payment (or prepaid key) arrives, uncovered mints are refused unpaid before verification with live_examples (mints covered right now) and the free /v1/radar to pick from, and eligibility is checked again after verification. Stale feeds and malformed mints are refused before payment. The existing full card remains $0.025 and the existing free routes remain free.

Start here

  1. Choose one to five pump.fun mints from your own workflow. Call GET /v1/check/coverage?mints=MINT1,MINT2 for free to see availability and exact prices. Coverage is also checked automatically by the paid route, so a separate preflight call is optional.
  2. Call GET /v1/check/mint/MINT ($0.005) or GET /v1/check/watchlist?mints=MINT1,MINT2 ($0.01 total). A payable request returns the standard x402 402 challenge. Pay once to receive the compact check and a watch cursor.
  3. Poll GET /v1/check/watch?mints=MINT1,MINT2&cursor=CURSOR every 15 seconds or slower for free. When data.has_new_events is true, pass the same cursor to the paid check for details and an updated cursor. Passing an unchanged cursor directly to the paid check also returns an unpaid no_new_events result.

Only these events advance a watch cursor: creator_sold, bundle_dumped, curve_drained, pool_drained, curve_closed_low, migrate_below_grad, and curve completion/migration/graduation transitions. Ordinary trades, ages and changing totals do not. Cursors bind to the exact mint set and expire after 24 hours. Calling the free watch without a cursor initializes a baseline; omit the cursor from a paid check when you want the current measurements regardless of new events.

The five-second eligibility limit is measured at result generation, before settlement. Settlement and network transit can add delay. meta.freshness_checked_at and feed_lag_at_generation_s record the eligibility check; result_age_s, feed_lag_s and is_stale are aged when the result is sent, including time spent settling. Compare t with your own clock after receipt; this is not a guarantee of delivery within five seconds.

Try it without a wallet

A shared public trial key is published at /llms.txt (section "Try it without a wallet") and in /pricing. Send it as X-API-Key on any route priced $0.01 or less (launches, the compact check, the watchlist, the verdict, the screened list, the outcome labels). It has tiny daily caps ($0.50 a day shared, $0.02 a day per caller IP) and answers carry X-Tape-Trial: 1 and a meta.trial block with the remaining amount; trial calls are never counted as payments. For your own budget buy a prepaid key: GET /v1/keys/trial ($0.10 of credit) or GET /v1/keys/new ($2.00).

Drop-in pre-buy filter for a self-built pump.fun bot: examples/python/pumpfun_bot_prebuy_filter.py asks the free coverage check, then the $0.01 verdict with your key, and returns (ok, reason); it fails open on any error so a slow check never blocks your bot. Outcome labels for grading your own signals: GET /v1/outcome/{mint} ($0.001) and GET /v1/outcome/batch?mints= (up to 40, $0.01) return, by the fixed public rule, whether a mint rugged or truly graduated and when. Free feeds: Atom feeds of graduation odds, Robinhood Chain ring flags and daily statistics at /feeds, usable by any RSS bot.

Pay on Solana, Base, Robinhood Chain or X Layer

Every paid route offers four x402 accepts entries: USDC on Solana mainnet and USDC on Base eip155:8453 (both settled by the Coinbase CDP facilitator since 2026-09-23), USDG on Robinhood Chain eip155:4663 (facilitator.naven.network) and USD₮0 on X Layer eip155:196 (PayAI facilitator). USDG and USD₮0 are not default assets in the x402 SDKs, so allow them in your client's spend controls if you want those entries. Clients that pay the first entry can reorder the quote with ?rail=solana|base|usdg|xlayer (or the X-Rail header); the order is Solana, Base, USDG, USD₮0 by default.

In a browser, no client needed: open any paid URL (for example a mint from the free radar) in a browser with a Solana wallet such as Phantom or Solflare; the page asks the wallet to pay the exact amount and then shows the result. With an EVM wallet (Coinbase Wallet, MetaMask) add ?rail=base to the URL. Agents get the JSON 402 with the PAYMENT-REQUIRED header instead.

Prepaid key, no client after one payment: GET https://api.loopholetape.com/v1/keys/new costs $2.00 once (x402 by header, or in a browser with a wallet) and returns a key lt_... holding $2.00 of credit. Then send X-API-Key: lt_... on any paid HTTP route and the route's price is deducted; GET /v1/keys/balance with the header is free. Credit does not expire; the key is a bearer secret. The same header on POST /mcp pays the MCP tools: set X-API-Key: lt_... in your MCP client's config for this server (clients that cannot sign x402 can still send a static header).

The bundled example buyers pay with USDC on Solana. Any generic x402 v2 client with an EVM signer pays the same prices with USDC on Base: for example @x402/fetch with @x402/evm (viem account) or Python x402[httpx,evm]. Check accepts[] for the eip155:8453 entry; its payTo and asset (native USDC) are fixed and published in /.well-known/x402.

Budgeted buyers: Python and TypeScript

Whole agent loop in one file: examples/python/agent_loop.py buys a $2 prepaid key once with x402, then polls GET /v1/launches/since every 30 s and buys a compact check for every launch whose calibrated P(true graduation) clears a threshold, all with a plain X-API-Key header. TAPE_PAYER_KEYPAIR=~/.config/solana/payer.json python examples/python/agent_loop.py --every 30 --grad-min 0.3 --check-max 20

The reference buyers work with HTTP and MCP. They check the recipient, Solana mainnet, USDC, exact scheme, resource, advertised product price and your budget before signing. A state file retains the original signed payment before transmission; timeouts reuse that payment. Neither client loads a wallet for unavailable or unchanged results. Choose a language below and run its setup from the repository root.

Python 3.12+ (Linux/macOS):

cd examples/python
python3 -m venv .venv
.venv/bin/pip install -r requirements-client.txt
.venv/bin/python risk_client.py "$MINTS" --coverage
# The next command authorizes real payments from YOUR funded wallet, within $0.05 total.
TAPE_PAYER_KEYPAIR=/absolute/path/to/payer.json .venv/bin/python risk_client.py "$MINTS" --budget 0.05 --polls 20

TypeScript, Node.js 24+:

cd examples/typescript
npm ci --ignore-scripts
node risk_client.ts "$MINTS" --coverage --state buyer-state-ts.json
TAPE_PAYER_KEYPAIR=/absolute/path/to/payer.json node risk_client.ts "$MINTS" --budget 0.05 --polls 20 --state buyer-state-ts.json

Set MINTS to one mint or a comma-separated list of at most five. Add --transport mcp and use a separate --state file to use the same workflow through MCP. --polls 20 checks once, then makes up to 20 free availability polls at 15-second intervals; it buys an update only when a watched event changes. Omit --polls for a single check. --max-per-call defaults to $0.01 and the single-mint route is capped at $0.005.

Keep each buyer-state*.json file private and use one active process per file. The cumulative budget belongs to that file and counts authorized attempts conservatively, including attempts that did not settle; rerunning does not reset it. A mint-set or transport change requires a separate state file. Both clients stop on an uncertain outcome rather than signing again. TypeScript uses an exclusive .lock file; after an abrupt crash, confirm that its recorded PID has exited before removing only that lock and resuming with the original JSON state. Python releases its process lock automatically.

Exact retries on the same transport and original arguments recover the original result and receipt for ten minutes. Replays carry meta.replayed, meta.result_age_s and aged freshness metadata; they do not contain newly measured data. The optional x402 payment identifier is supported, but an identifier alone cannot retrieve a result. Altering metadata or arguments on an already-used transaction returns a conflict. If you receive payment_outcome_unknown with charged: null, or the retry window has elapsed, reconcile the original transaction before authorizing another payment.

Both examples include pinned dependency files for installation.

What it is

Live pump.fun / PumpSwap analytics derived from public on-chain data. Pay per request in USDC on Solana through x402. See the documented routes for coverage and freshness.

Since v0.3.0 the same service also covers Robinhood Chain (Pons V2 launchpad): launch feed and per-curve structure cards derived from public on-chain activity (exact reserves, raised/progress/price multiple, the launch's trade fee and creator tax in basis points, snipe-tax activity, holder concentration, same-block direct-buy cluster share, direct vs terminal route mix, new-buyer flow, lifecycle).

Robinhood Chain two- and six-second flow counts expire against the response timestamp, even when no new trade arrives. Check meta.is_stale before interpreting zero activity; field definitions explain timestamp and coverage semantics.

Routes

RoutePrice (USDC)What you get
GET /v1/check/mint/{mint}$0.005Compact observed-risk check for one fully covered pump.fun mint
GET /v1/check/watchlist?mints=...$0.01 totalCompact checks for up to five fully covered caller-selected mints
GET /v1/check/batch?mints=...$0.04 totalCompact checks for up to 25 mints in one call (uncovered mints reported per item)
GET /v1/launches/screened?limit=$0.01The newest covered pump.fun launches without an avoid verdict, ranked by calibrated true-graduation odds, each with its verdict word and reasons
GET /v1/mint/{mint}/holders?limit=$0.01Current holder table of one covered mint: shares, first buys, observed labels (creator, create-slot buyers, leaderboard classes), concentration summary
GET /v1/mint/{mint}/flow?since=&limit=$0.01 a pageEvery bonding-curve buy and sell of one covered mint in slot order with running net SOL, 60 s / 5 min summaries and a cursor
GET /v1/verdict/{mint}$0.01One word (avoid / caution / no_flags_observed) by a fixed public rule, the calibrated odds, up to three observed reasons, and a free shareable card page (/verdict/{token}, HMAC-signed)
GET /v1/calibrationfreeValidation tables behind the probabilities, plus the production block: the rug number is recalibrated on the radar's own live calls (before/after tables; it is withdrawn if live calibration error exceeds 0.05)
GET /v1/radar, page /radarfreeLive radar: the ten covered mints under 30 s with the highest calibrated rug-within-300s probability and the ten under 15 min with the highest true-graduation probability, each with the paid check URL; refreshed every 5 s
GET /v1/rhc/sample/launches, page /rhcfreeRobinhood Chain (Pons V2): five launch cards delayed 5 minutes in the exact paid shape, each with its ring block and the ring rule, and a page with the live regime meter (launches/h, graduation rate, creator-tax split)
GET /v1/keys/new$2.00Buy a prepaid key: $2.00 of credit spent by sending X-API-Key on any paid HTTP route (activated by the same request's settlement)
GET /v1/keys/trial$0.10The same prepaid key with $0.10 of credit (about twenty checks) for trying the header path with a small budget
GET /v1/keys/balancefreeStatus and remaining credit of a prepaid key (X-API-Key header)
GET /v1/datasets, /v1/datasets/sample/pumpfun_launches.csv, /v1/datasets/schema/pumpfun_launchesfreeBulk datasets index, a 200-row sample and the column schema
GET /v1/datasets/pumpfun_launches/{day}$5.00One UTC day of pump.fun launches with outcomes and early features as a zstd parquet (about 15k-35k rows); the MCP tool buy_dataset returns a one-hour download URL instead
GET /v1/launches/since?since=&limit=$0.001Polling feed for bots: every covered launch created after the cursor, oldest first, with its verdict word, the calibrated probabilities and the check URL; filters no_avoid=1 and min_p_grad=; returns next_cursor
/stats/pumpfun, /stats/pons (JSON at /v1/stats/*)freeDaily launch statistics from our own capture: true graduation rate, the instant-bundle share of graduations, organic rate, ticker-copy counts; Robinhood Chain launches, graduations, creator-tax and stock-token-quote shares
GET /go/{venue}/{mint}freePlain redirects to pump.fun, DexScreener, Jupiter, GMGN, Axiom or Photon for a mint (no referral code, no fee)
GET /v1/track-record, page /track-recordfreeDaily scorecard of the radar's own calls against outcomes: rug flags rugged within 300 s, graduation flags graduated, base rates
GET /v1/check/coverage?mints=..., GET /v1/check/watch?mints=...&cursor=...freeCoverage/prices and availability of new documented watch events
GET /v1/mint/{mint}$0.025Risk card for one pump.fun mint (live microstructure or thin on-chain card)
GET /v1/launches/recent$0.01Recent pump.fun launches with compact risk labels
GET /v1/rugs/recent$0.02Recent rug / close flags with mechanism labels
GET /v1/graduations/recent$0.02Recent migrations labelled true graduation vs below-threshold close
GET /v1/creator/{address}$0.02Creator reputation over our full launch history
GET /v1/wallet/{address}$0.02Wallet class from our rolling PnL leaderboard + live holdings
GET /v1/rhc/launches/recent$0.01Robinhood Chain (Pons V2) recent launches with holder / route / cluster structure, and the ring screen: ring_only=true returns only curves matching the coordinated-cluster shape (8+ buyers, same-block direct-buy cluster share ≥ 0.5) with the rule and its measured base rates beside the items
GET /v1/rhc/curve/{address}/trades?since=&limit=$0.02 a pageRobinhood Chain (Pons V2): every buy, sell and fee-sweep buyback of one curve with the reserves after it (rebuilt to the wei with the exact four-word curve model), marginal price, fee, snipe tax and creator tax; free 5-minute-delayed sample at GET /v1/rhc/sample/trades?address=
GET /v1/rhc/curve/{address}$0.02Robinhood Chain (Pons V2) curve structure card for one curve or token address, with a trade_cost block: for a 0.01 and a 0.05 ETH clip at the reserves now, the fill over the marginal price, the immediate round trip and the price rise needed to break even, through the exact constant product with the 1% fee and the launch's creator tax on both legs (null with a reason on non-ETH pairs, finished curves or inside the snipe window)
/, /health, /v1/market/regime, /v1/sample/launches, /v1/sample/rugs, /v1/sample/mint, /v1/x402/resources, /v1/labels, /v1/rhc/regime, /pricing, /about, /transparency, /.well-known/x402, /openapi.json, /llms.txt, /mcp, /terms, /.well-known/agent-card.json, /a2a, /.well-known/mcp-server-card, /.well-known/agents.json, /agents.txt, /.well-known/api-catalogfreeindex, health, regime meters, delayed samples, discovery documents, labels, MCP, prices, seller identity, settlement record, terms

GET /about (aliases /about.json, /team, /contact) answers who sells this: the project identity and its role mailbox, this repository, the signed proof of control of both payment addresses (check each signature against accepts[].payTo in any 402 challenge), where the data comes from, and the plain fact that there is no token, presale or fundraising. HTML for browsers, the same record as JSON for agents.

The same identity resolves as a DID: did:web:api.loopholetape.com, document at GET /.well-known/did.json. Both payout accounts appear as CAIP-10 blockchainAccountId verification methods (solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp:... and eip155:8453:0x...), and the signatures that bind them to the origin are in the document's extension block, so a directory can verify the seller rather than trust it.

GET /transparency (aliases /transparency.json, /.well-known/transparency.json, /disclosures, /trust) is the seller's own settlement record: how many outside settlements there have been and for how much, split by payer class (a crawler that pays dozens of sellers in one sweep is counted apart from a customer), by network, with our own registration payments excluded and counted separately; the charging and replay policy; every named reason a payment can be refused (reason and hint come back on the 402, so a refusal says what to correct); known incidents; and both payout addresses, so the counts can be checked against the chain instead of believed. No payer wallet or purchase is published.

Every paid response is self-describing: schema_version, generated_at, coverage, and meta with freshness, source and related resources. The full /v1/mint/{mint} card supports both full live coverage and thin history/on-chain coverage. New compact checks require full coverage; an unavailable mint makes the entire requested batch unpaid. Label semantics: /v1/labels. Seven-day base rates are included in the existing full card.

Discovery for agents

An agent skill describing the whole paid workflow is served at /skills/pumpfun-risk-check/SKILL.md and published on Smithery as loopholetape/pumpfun-risk-check; the MCP server is listed as loopholetape/loophole-tape. Agent directories can read the A2A agent card at /.well-known/agent-card.json (the /a2a JSON-RPC endpoint answers message/send with the catalog, payment terms and the exact call for a named skill or mint), the MCP server card at /.well-known/mcp-server-card, agents.json flows at /.well-known/agents.json, agents.txt, and the RFC 9727 catalog at /.well-known/api-catalog.

Use the live documents for current capabilities, schemas and prices:

MCP registry name: io.github.gosadu/loophole-tape. Connect to https://api.loopholetape.com/mcp.

Pay from Python (reference client)

For other routes, use the generic HTTP buyer from examples/python after the Python setup above:

TAPE_PAYER_KEYPAIR=~/.config/solana/payer.json .venv/bin/python pay_client.py https://api.loopholetape.com "/v1/launches/recent?limit=5"

The reference clients register a payment guard: they sign only for this API's recipient, Solana mainnet and USDC, at or under the route's price (cap TAPE_MAX_USD_PER_CALL, default $0.03); anything else is refused. Any x402 v2 client works the same way: call the route, read the PAYMENT-REQUIRED header (the 402 body repeats it as JSON with a how field), sign the USDC transfer, retry with PAYMENT-SIGNATURE; the PAYMENT-RESPONSE header carries the settlement signature. TypeScript: @x402/fetch + @x402/svm. AgentKit's discover_x402_services finds the routes.

Routers that cannot fill a path template can use the query form of the per-id routes (/v1/verdict?mint=, /v1/check/mint?mint=, /v1/mint?mint=, /v1/creator?address=, /v1/wallet?address=, /v1/rhc/curve?address=): quoted and served for the id named, at the path form's price; the bare form without an id answers a 402 that names the required parameter and refuses a payment without one.

MCP (for agents that call tools)

MCP server over streamable HTTP at https://api.loopholetape.com/mcp (no trailing slash needed; CORS preflight and plain JSON accepted). 28 tools: eight free, twenty paid (tools that declare an outputSchema publish anyOf[success, x402 PaymentRequired, refusal], so both the TypeScript and Python x402 clients pay them) (the verdict tool, $0.01, returns the verdict object with its share link). The compact-check tools are check_coverage and watchlist_updates (free), check_pumpfun_risk ($0.005) and check_watchlist_risk ($0.01 total). MCP mint lists are JSON arrays; a single check takes mint. The new paid tools publish typed outputSchema and return the same data as HTTP.

Existing tools: catalog, health, market_regime, sample_mint, radar, rhc_regime (free); mint_risk_card ($0.025), recent_launches ($0.01), recent_rugs, recent_graduations, creator_reputation, wallet_profile, rhc_curve_card ($0.02), buy_api_key ($2.00 once, a prepaid key for the HTTP routes and these tools), buy_trial_key ($0.10), buy_dataset ($5.00 per day file), launches_since ($0.001 per poll), and rhc_recent_launches ($0.01).

Paid tools first return an isError payment-required result. Retry with the signed payload in _meta["x402/payment"]; settlement returns in _meta["x402/payment-response"]. Invalid inputs and unavailable compact checks return structured errors without requesting payment. Use the budgeted buyers with --transport mcp. For other tools, use the generic MCP buyer; from examples/python, install its extra dependency with .venv/bin/pip install "mcp<2".

Terms, in one line

Statistics derived from public on-chain data; labels are heuristics, not guarantees; nothing here is financial advice; sold as-is per request; no refunds for empty results; do not resell raw responses. Full text at /terms.