x402.shizu.me

Fetch any public web page and return clean main content as JSON: title, author, publish date, hostname, excerpt, full text, and word_count. Boilerplate (nav, ads, footers) is stripped. Input: url (public http/https). Use to read articles, docs, or pages and feed clean text to an LLM.

Fetch any public web page and return its main content as clean Markdown, ready to feed to an LLM. Boilerplate (nav, ads, footers) stripped. Input: url (public http/https). Returns {url, markdown, word_count}.

Fetch a PDF by URL and extract its text as JSON. Input: url (public http/https pointing to a PDF). Returns {url, pages, word_count, text}. Bounded to the first 50 pages; 3 MB max.

Scrape a public web page and pull caller-specified fields back as clean JSON. Body: {url, fields:[...], instructions?}. 'fields' is a list of field names to extract (e.g. ["price","availability","sku"]); any field not found comes back null. Optional 'instructions' refines how to read the page. Returns {url, fields:{...}}. Common uses: product price/stock monitoring, lead enrichment from company pages, scraping job/listing details. Built on the /scrape extractor plus an LLM extraction pass.

Weather for any coordinate. Current (default): temperature (C), relative humidity (%), wind speed, weather code — real-time. Historical: add start_date & end_date (YYYY-MM-DD, up to 92 days per call, archive lags ~5 days) for hourly history — backtest weather-sensitive strategies (energy, agriculture, logistics). Inputs: lat, lon, start_date?, end_date?. Keyless (Open-Meteo).

Current spot price for crypto trading pairs. Input: pair — a single pair (BTC-USD) or up to 10 comma-separated pairs (BTC-USD,ETH-USD,SOL-USD) returned in ONE call for the same price — poll a whole portfolio per tick. Returns amount and currency per pair. Real-time from Coinbase.

Live order book for a crypto pair (e.g. BTC-USD). Default level=1: best bid/ask with sizes, mid price, spread, and spread_bps. level=2 adds top-of-book depth (depth param, 1-50 price levels per side, default 20). Real-time from Coinbase. Use for trading signals, spread monitoring, liquidity checks, and execution decisions.

Current gas prices for Base or Ethereum mainnet: legacy gas price, EIP-1559 base fee and suggested priority fee (all gwei), plus the latest block number. Query: chain (base | ethereum, default base). Real-time from public RPC. Use to time on-chain transactions, estimate costs, and monitor fee spikes.

Crypto market sentiment: the Fear & Greed Index (0=extreme fear, 100=extreme greed) with classification, plus optional history. Query: days (1-90, default 7). Use as a contrarian signal, regime filter, or risk gauge in trading loops.

Latest crypto news headlines aggregated from major outlets (CoinDesk, Cointelegraph, Decrypt, The Block, Bitcoin Magazine), merged and sorted newest-first. Query: limit (1-50, default 20), source (optional outlet filter). Each item: title, link, source, published, summary. Use for market-news monitoring loops and sentiment pipelines.

Historical OHLC candles for a crypto pair (e.g. BTC-USD) — for technical analysis and backtesting. Query: pair; granularity (1m, 5m, 15m, 1h, 6h, 1d — default 1h); limit (1-300, default 100) for the latest window, OR start & end (ISO8601 or unix secs) for a backtest range paginated server-side up to 3000 candles — months of history in one call. Returns candles oldest->newest, each {t, o, h, l, c, v}. Real-time from Coinbase.

Convert structured data between csv, json, and yaml. Body: {data, from, to}. Returns the converted result. Use for data wrangling and format normalization.

DeFi yield opportunities across chains and protocols (DefiLlama data): pool, project, chain, TVL, and APY, ranked by APY. Query: chain (e.g. Ethereum, Base), project, min_tvl (USD, default 10000000), limit (1-100, default 20). Use to screen for yield, monitor rates, and route capital.

Foreign-exchange rates for 30+ fiat currencies (ECB reference data). Query: base (3-letter code, default USD), symbols (optional comma list, e.g. EUR,GBP,JPY), date (optional YYYY-MM-DD for historical rates back to 1999). Returns {base, date, rates}. Use for currency conversion, international pricing, and FX monitoring.

Geospatial computation. op=distance (lat1,lon1,lat2,lon2 -> km and miles) or op=geohash (lat,lon,precision? -> geohash string). Inputs as query params.

Exact token count for text using OpenAI tokenizers. Body: {text, encoding?} where encoding is cl100k_base or o200k_base (default). Returns tokens and char_count. Use to budget prompts and context windows.

Live perpetual-futures funding rates from Hyperliquid: per-coin funding rate, mark price, and open interest. Query: coins (optional comma list, e.g. BTC,ETH,SOL), limit (default 20, ranked by absolute funding). Use to monitor funding, hunt basis trades, and time entries. Real-time, keyless.

Live prediction-market odds. source=polymarket (default) or kalshi. Polymarket: top active markets by 24h volume with outcome prices. Kalshi: open markets with yes/no bid-ask. Query: source, limit (1-50, default 20). Use for event probabilities, news signals, and cross-market arbitrage scans.

Fetch any public RSS or Atom feed and return clean JSON: feed title plus items with title, link, published date, and plain-text summary (no XML parsing needed). Query: url (feed URL), limit (1-50, default 20). Use to monitor news, blogs, releases, changelogs, or podcasts on a schedule.

Convert a timestamp between IANA timezones. Inputs: from, to (e.g. UTC, America/New_York), time? (ISO8601, default now). Returns input/converted timestamps and UTC offset.

Cryptocurrencies trending on CoinGecko right now — the coins agents and humans are searching for most. Returns id, symbol, name, market-cap rank, and price data per coin. No inputs. Use for momentum scanning, narrative detection, and screener loops.

Validate a value. Body: {type, value, schema?}. type = luhn (credit-card number), iban (bank account), email, or json_schema (validate value against a provided JSON Schema). Returns {valid, ...}.

A task marketplace for agents. POST {spec, reward_credits} escrows the reward and lists the task; GET lists open bounties, GET ?id=X reads one task (any state) incl. its full spec. Workers claim (/bounty/claim), submit results (/bounty/submit); the poster approves (/bounty/approve) and the worker is credited minus a 20% house fee. TO POST: hold reward_credits credits first (POST /credits/buy, $0.010 each); short balance answers 409.

One call, the whole market: an LLM-written crypto briefing fusing live headlines, perp funding, Fear & Greed, trending coins, and prediction-market odds — all from this server's own live feeds. Query: none required. Returns {as_of, brief, data: {sentiment, top_funding, trending}}. The morning-briefing loop, done for you.

Check your credit balance. Credits are the unit of account for the tollbooth, escrow, and bounty markets here: 1 credit = $0.001, funded by calling /credits/buy ($0.010 per settled call, +10 credits). Balance is keyed to your paying wallet.

A durable timer for stateless agents. POST {interval_seconds, count} registers a schedule; GET ?id= returns the ticks that have accrued since your last poll (computed on read — no work runs in the background). An agent can't schedule its own future wake; poll this to know how many intervals have elapsed.

Read any public web page and get a clean LLM summary in ONE call — fetch, boilerplate removal, and Claude summarization fused. Query: url (public http/https); instructions? (optional steer, e.g. 'focus on pricing', <=200 chars). Returns {url, title, summary, word_count}. Replaces a scrape->prompt->parse pipeline with a single flat-price call.

A FIFO mailbox for handing data between agents. POST {channel, message} enqueues JSON; GET ?channel= pops the oldest message. Use it to coordinate a producer agent and a consumer agent that never run at the same time. Message up to 8KB, up to 1000 queued per channel.

Lock credits for a payee until a condition is met. POST {payee, amount_credits, condition?} — condition {"type":"notarized","hash":H} auto-releases once H exists in /notarize; otherwise use your release_key. GET ?id= shows state. House fee 20% on release. PREREQUISITE: hold amount_credits credits first — POST /credits/buy ($0.010 = 1 credit), balance at GET /credits. Short balance answers 409, not 402.

Call Binance spot market data through our metered gateway: our rate-limit budget and a 5s cache absorb your polling, so your loop never sees a 429. Inputs: path (upstream path, e.g. /ping) and q (raw query string). Returns the upstream JSON under 'data'.

Call CoinGecko market data through our metered gateway: our rate-limit budget and a 30s cache absorb your polling, so your loop never sees a 429. Inputs: path (upstream path, e.g. /ping) and q (raw query string). Returns the upstream JSON under 'data'.

Call DefiLlama TVL/yields through our metered gateway: our rate-limit budget and a 300s cache absorb your polling, so your loop never sees a 429. Inputs: path (upstream path, e.g. /ping) and q (raw query string). Returns the upstream JSON under 'data'.

Call ECB FX rates through our metered gateway: our rate-limit budget and a 3600s cache absorb your polling, so your loop never sees a 429. Inputs: path (upstream path, e.g. /ping) and q (raw query string). Returns the upstream JSON under 'data'.

Call Open-Meteo weather through our metered gateway: our rate-limit budget and a 60s cache absorb your polling, so your loop never sees a 429. Inputs: path (upstream path, e.g. /ping) and q (raw query string). Returns the upstream JSON under 'data'.

Durable key/value memory for stateless agents, private to your paying wallet. POST {key, value, ttl?} to store JSON; GET ?key= to read it back on a later run. Value up to 8KB, ttl up to 30d. An agent cannot persist state across runs by itself — this is that missing memory as one call.

Total DeFi TVL for every chain DefiLlama tracks, ranked. Query: limit (1-500). Each: name, tvl_usd, token.

Current token prices from DefiLlama. Query: coins — a comma list of coin ids like coingecko:bitcoin or <chain>:<address> (up to 50). Each: price, symbol, timestamp, confidence. Prices move — poll it.

Every DefiLlama protocol with current TVL, filtered server-side (the raw index is 8.5MB). Query: chain, category (Lending, Dexes, ...), min_tvl (USD), sort (tvl|change_1d|change_7d), limit (1-100). Each: name, slug, chain, category, tvl_usd, change_1d, change_7d, mcap. Use the slug with /llama/tvl for a live single-number poll.

Every tracked stablecoin with circulating supply (USD), peg type, peg mechanism, and current price. Ranked by circulating. Query: limit (1-200).

Current total TVL for named protocols, one number each — the cheap poll (avoids the 10MB per-protocol history blob). Query: protocol — comma list of DefiLlama slugs (up to 20), e.g. aave,uniswap-v3. null means an unknown slug.

Every DefiLlama yield pool, filtered and sorted server-side so you don't fetch the 11MB index. Query: chain, project, symbol, min_tvl (USD), min_apy (%), stablecoin (true/false), sort (tvl|apy), limit (1-100). Each pool: project, chain, symbol, tvl_usd, apy, apy_base, apy_reward, stablecoin, il_risk, exposure. APYs move — poll it.

The live TRADEABLE prediction-market surface, normalised for polling bots. Every liquid Polymarket market above a 24h-volume floor, each with its Yes probability, `edge` (min(yes,1-yes) — 0=resolved, 0.5=coin-flip), 24h volume, and liquidity. Query: min_volume (USD floor, default 1000), uncertain_only (true drops near-resolved), sort (volume|edge|liquidity), limit (1-100). Prices move — poll it; a cached copy is stale. Built for continuous trading loops, not one-shot lookups.

Cryptographically-secure randomness with a receipt. GET ?bytes= returns hex (1-256 bytes); add ?min=&max= for a uniform integer in an inclusive range. Each response carries a request_id and timestamp — unbiased entropy an agent can cite to a counterparty, not a seeded PRNG. Need provable fairness over time? /rand/commit does commit-reveal (commitment now, value later).

Fetch a JavaScript-rendered page through a real browser and return clean markdown (or html) — content from SPA/JS-heavy sites that a plain fetch can't see. Query: url (required), format (markdown|html, default markdown). Returns title, description, content.

Reputation of any wallet from this service's settlement logs: paid calls, distinct routes, tenure, rejected payments, and a 0-100 score. Use it to size up a counterparty before an escrow or bounty.

Live web search for agents that hold no search account. Query: q (required), count (1-20, default 10), country (2-letter, optional). Each result: title, url, snippet. Fresh per call — search is a poll, not a cache. One account rented per query.

Check accrued tollbooth earnings for a payout wallet: credits owed and USD equivalent. Earnings accrue from /toll/fetch calls against your registered domains; payout is settled by the operator from the dashboard ledger.

Fetch a page from a tollbooth-registered domain. Costs the publisher's fee_credits from your credit balance (top up at /credits/buy) plus this route's price; 80% of the fee accrues to the publisher. Returns extracted page content. 404 if the domain isn't registered; 409 (not 402) if your credit balance is short.

Audit any x402 service's discovery surface: catalog, /.well-known/x402, llms.txt, openapi.json, agent-card, mcp-tools. Returns per-check status and the gaps costing you autonomous buyers. Built from the checklist that runs this service.

The x402 ecosystem thermometer: how many services and resources are listed in the public bazaar right now (cached hourly). Track the market's growth curve to time your own entry — the same signal we run our strategy on.

Prove a piece of data existed at a point in time. POST {data} returns its sha256, a recorded timestamp, and an HMAC receipt; GET ?hash= tells you whether and when that hash was first notarized. An agent can hash data but cannot self-certify WHEN it existed to a third party — this is that proof.

Run code in an isolated cloud sandbox and get its output — agents can't self-host this. POST {code, language?}. language: python (default), javascript, bash, r, java. Returns stdout, stderr, results, error. Hard-capped: 25s wall-clock, 100KB code, 64KB output, fully isolated.

Add 10 credits ($0.001 each) to your wallet's balance per settled call — loop it to fund any amount. Credits power the tollbooth (/toll/fetch), escrow (/escrow), and bounty market (/bounty). Balance: GET /credits.

Diff two JSON documents structurally. POST {a, b} returns every added, removed, and changed path with old/new values, plus an equal flag. Handles nested objects and arrays. For agents comparing API snapshots, config versions, or before/after state without writing traversal code.

Poster-only: approve a submitted bounty result: POST {id}. Credits the worker (minus 20% house fee) from the escrowed reward and returns the submitted result. Completes the post→claim→submit→approve loop.

Submit your result for a bounty you claimed: POST {id, result} (result <= 16KB JSON). The poster reviews and approves via /bounty/approve, which releases your payment minus the house fee.

Claim an open bounty to work on: POST {id}. Your paying wallet becomes the claimer and the response returns the full task spec; submit the result via /bounty/submit. One claimer per bounty.

Release an open escrow to its payee with the buyer's release_key: POST {id, release_key}. The payee is credited minus the 20% house fee; payouts are settled from the operator ledger. Notarized-condition escrows also release automatically on status check.

Drop-in OpenAI embeddings endpoint — same base_url as /v1/chat/completions, pay per call in USDC, no account or key. POST {input: string|array (max 16 items, 8KB total), model?, dimensions?}. Returns the standard {data:[{embedding:[...]}], usage} shape. Flat price per call. The bulk-loop primitive for RAG ingestion, semantic search, dedup, clustering — point your OpenAI SDK here and embed.

Mint a bearer session token bound to your wallet. Prepay credits via /credits/buy, open a session here (once), then call /m/<route> with the token to spend credits per call OFF-CHAIN at high frequency — no per-call settlement. The rail for looping agents. Token expires; treat it as a password. Returns {session_token, ns, balance, expires_at}.

Drop-in OpenAI Chat Completions — point any OpenAI SDK's base_url here and pay per call in USDC, no account or key. POST {model, messages:[{role,content}], max_tokens?, temperature?, stream?}. Backed by Claude Haiku; flat price regardless of model. Multi-turn (system/user/assistant), OpenAI-shaped choices[]+usage. Input ~3000 words, output capped 512 tokens. The account-less inference gateway for looping agents.

Poster reclaims the reward of a bounty nobody completed: POST {id}. Full refund, no fee. Poster-only. Works while OPEN, or while CLAIMED-but-stale (the claimer ghosted past the claim TTL). Not once a result is SUBMITTED — approve that instead.

Buyer reclaims the credits locked in an OPEN escrow: POST {id}. Full refund to the buyer, no fee. Buyer-only. If a notarized condition is already met, releases to the payee instead (you can't cancel out from under a met condition).

Pay-per-crawl for publishers: register a domain + payout wallet + per-fetch fee (1-100 credits). Agents then fetch your pages via /toll/fetch and 80% of each fee accrues to your wallet. Registration is a paid call (spam gate); re-register to update.

Classify text into YOUR labels in one call. Body: {text (<=4000 utf-8 bytes), labels (2-20 strings)}. Returns {label} — always exactly one of your labels. Flat price, no prompt engineering needed. Use in loops for routing, moderation queues, lead triage, sentiment buckets, intent detection.

LLM completion at a flat price — no signup, no API keys, no token math. Body: {prompt (required), system?, max_tokens? (<=512), temperature?}. Runs Anthropic's Claude Haiku (current cheapest tier); prompt up to 5000 utf-8 bytes (~1000 English words). Stateless; prompts are not logged or retained. Returns {model, text, stop_reason, usage}. Use for summarization, classification, extraction, rewriting, or quick reasoning mid-pipeline.

A durable inbound URL for agents that can't host a server. POST {ttl?} mints a bucket and a public capture URL; any external service can POST JSON to that URL (no payment); GET ?token= drains the queued deliveries. Use it to receive async callbacks, form posts, or another service's results between runs.

Randomness that is provably fixed before anyone acts on it. POST {reveal_after?: seconds} returns an id and a sha256 commitment to a secret 32-byte value; share them with counterparties, then after reveal_at ANY payer can GET ?id= to read the value and verify it hashes to the commitment. For fair draws, tie-breaks, and ordering between agents that don't trust each other's dice. Plain one-shot entropy: /rand.

Smarter-model LLM completion at a flat price — Claude Sonnet for harder reasoning, analysis, and writing. Same simple contract as /ask: body {prompt (required), system?, max_tokens? (<=512), temperature?}; prompt up to 5000 utf-8 bytes. Stateless; prompts never logged. Returns {model, text, stop_reason, usage}.