MCP server exposing 69 tools - LLM completion, read-only EVM tooling and pure crypto utilities. initialize and tools/list are free so any client or directory can enumerate the tools; only tools/call is charged, a flat $0.001 per call.

Turns a document into clean Markdown in about a second: pass an https URL or base64 of a PDF, Word, Excel, PowerPoint, HTML, CSV or XML file and get the text back with its headings, tables and page markers, plus the token count. It reads the text layer, so a scanned page with no text layer comes back without text. You are only charged if the Markdown is delivered. No API key, no account.

Generates a 1024x1024 JPEG from a text prompt with FLUX.1 [schnell] in about 2 seconds. Returns a hosted image URL valid for 24 hours that you can hand straight to the next tool; add inline=true to also get the base64. Choose 1-8 diffusion steps (default 4). You are only charged if the image is delivered. No API key, no account.

Describes what is in an image in about one second: pass an https URL or base64 and ask anything about it, from a one-line caption to reading a sign. Pairs with the rest of the chain: describe an image, translate the description, then have it read aloud. You are only charged if the description is delivered. No API key, no account.

Edits an image you provide instead of generating one from scratch: pass the image and a black-and-white mask, and Stable Diffusion inpainting repaints only the white area following your prompt. Image and mask go as https URLs or as base64, up to 10 MB each. Returns a hosted image URL valid for 24 hours. You are only charged if the edit is delivered. No API key, no account.

Generates an illustration from a text prompt with DreamShaper 8 LCM in about 2 seconds: painterly and character art rather than photoreal. Takes a seed for a repeatable result, a negative prompt, and your own width and height. Returns a hosted image URL valid for 24 hours; add inline=true to also get the base64. You are only charged if the image is delivered. No API key, no account.

Turns text into natural speech in English, Spanish, French, Chinese, Japanese or Korean in one or two seconds: Deepgram Aura-2 for English and Spanish, MeloTTS for the rest. Returns a hosted audio URL valid for 24 hours (MP3 for English and Spanish, WAV otherwise) plus its size; add inline=true to also get the base64. Up to 2000 characters per call. You are only charged if the audio is delivered. No API key, no account.

Transcribes an audio file with Whisper large v3 turbo: pass an https URL (mp3, wav, flac, m4a, ogg, up to 10 MB) or base64 audio in a POST. Returns the text, the detected language with its probability, the duration, and timed segments; add words=true for word-level timestamps. Around 100 languages. You are only charged if the transcript is delivered. No API key, no account.

Translates text between about 100 languages with Meta M2M100 1.2B: pass the text and the target language (ISO code like es, fr, de, zh, ja, or its English name); the source language is detected if omitted. Long texts are split by sentence, so paragraphs keep their order and line breaks. Up to 2000 characters per call. Pairs with speech-to-text: translate a transcript, then read it aloud. You are only charged if the translation is delivered. No API key, no account.

One piece of practical advice about work, life or money, in English or Spanish. Short, concrete and usable - not a fortune cookie. Filterable by topic, with an id to avoid repeats and a seed for reproducibility.

One curious but true fact, in English or Spanish, filtered by topic: science, computing, history or nature. Takes parameters, returns an id so you can avoid repeats across a session, and accepts a seed for a reproducible answer.

One two-sentence horror story, in English or Spanish, by theme: domestic, technological, cosmic or folk. Returned SPLIT into setup and payoff as well as joined, so a narrating agent can deliver the first line, pause, and then land the second. Takes parameters, returns an id to avoid repeats, and accepts a seed for a reproducible answer.

One random joke, in English or Spanish, filtered by category: programming, crypto or general. Unlike the usual one-liner endpoints this one takes parameters, returns an id so you can exclude what you have already seen, and accepts a seed for a reproducible answer.

One riddle by difficulty, in English or Spanish, with the answer AND the SHA-256 of the answer in lowercase. The hash is the point: an agent can check whether its guess is right without reading the solution first, which no other riddle endpoint lets you do.

One lighthearted developer roast aimed at a target YOU choose - javascript, python, css, git, regex, kubernetes, agile or legacy code - and at the severity you ask for. Every other roast endpoint returns something generic; this one lets a code-review agent roast the actual stack in front of it. Bilingual, seedable and repeat-free.

The word of the day with its definition, part of speech and an example sentence, in English or Spanish. Genuinely of the day: it is picked deterministically from the UTC date, so everyone gets the same word and you can reproduce any past day by passing it. The usual endpoints just return a random word and call it the word of the day.

Checks one crypto address against the OFAC SDN list published by the US Treasury, across 16 currencies including Ethereum, Bitcoin, Tron and USDT. A hit returns the actual list entry - sanctioned entity, its type and its sanctions programmes - not a risk score. Exact matching only, never fuzzy. Every answer carries the publication date of the list snapshot. No API key, no account.

Native coin balance (ETH, MATIC...) for any address, on Base, Ethereum, Polygon, Arbitrum or Optimism. Flat price, no rate limits, no API key. Via eth_getBalance.

Latest block height and timestamp of any supported chain, plus how old the chain head is in seconds. Tells an agent whether a node is in sync. Via eth_getBlockByNumber.

Runs any read-only contract call (eth_call) with your own calldata and returns the raw result, on five chains, with no API key and no rate limit. The escape hatch for anything the typed endpoints do not cover.

Live health of every supported chain in one call: chain id, block height and how stale the head is. Use it to pick a chain that is actually responding. Via eth_getBlockByNumber.

Resolves an ENS name to its address, or an address back to its primary ENS name, reading Ethereum mainnet directly. Names are what humans paste and addresses are what chains need.

How much of a token a spender is still allowed to move on an owner's behalf, and whether that approval is unlimited. The check to run before an approve, and the one that finds forgotten infinite approvals. Via eth_call (allowance).

ERC20 token balance for any wallet - USDC, USDT, DAI or any token - via balanceOf. Returns the raw amount and the decimals so you can format it. Via eth_call (balanceOf).

Estimates the gas a transaction would need and prices it at the current gas price, before you sign anything. Also the cheapest way to find out that a call would revert. Via eth_estimateGas.

Suggested maxFeePerGas and maxPriorityFeePerGas from the last 20 blocks of real fee history, at slow, normal and fast percentiles. Beats guessing a gas price and getting stuck in the mempool. Via eth_feeHistory.

Current gas price and base fee for any supported chain, in wei and gwei, plus the cost estimate of a plain transfer. For agents planning a transaction. Via eth_gasPrice.

Tells whether an address is a smart contract or a plain wallet, and how big its bytecode is. A cheap safety check before sending funds. Via eth_getCode.

Who owns a given ERC721 token id, plus its tokenURI and collection name in the same call. Answers is this NFT still theirs without an indexer or an API key. Via eth_call (ownerOf).

Next transaction nonce for any address, and whether the address is a contract. What an agent needs before signing a transaction. Via eth_getTransactionCount.

Native balance plus the balance of every token you list, on one chain, in a single paid call - each with symbol and decimals already applied. One request instead of one per token. Via eth_getBalance + eth_call.

Receipt of a transaction: whether it succeeded or reverted, gas used, fee actually paid, and how many logs it emitted. The call that answers did my transaction work. Via eth_getTransactionReceipt.

Reads a raw storage slot of any contract, decoded as hex, integer and address. How you inspect proxy implementation slots, owners and paused flags that have no public getter. Via eth_getStorageAt.

Name, symbol, decimals and total supply of any ERC20 contract in one call. Useful to label a token address before showing it to a user. Via eth_call.

Spot price of any ERC20 read straight from the deepest Uniswap V3 pool against USDC, on Base, Ethereum, Polygon, Arbitrum or Optimism. Returns the pool address, fee tier, liquidity, sqrtPriceX96, tick and block, so the number can be audited instead of trusted. No aggregator and no API key: the source is the chain itself. Via eth_call.

Pre-trade risk check for any ERC20 on Base, Ethereum, Polygon, Arbitrum or Optimism, straight from the deployed bytecode and live state. Detects mint, pause, blacklist, upgradeable proxy and mutable buy/sell tax powers, and reports whether ownership is renounced. Unlike LLM-written verdicts this returns the EVIDENCE - the exact selector found and its byte offset - so an agent can apply its own policy. Deterministic: same contract, same answer.

Full transaction details by hash, merged with its receipt: from, to, value, gas used, status and block. One call instead of two. Via eth_getTransactionByHash.

Converts an amount between any two of the ~30 currencies in the European Central Bank's official reference rates, on today's publication or on a past date within the last 90 days. Returns the rate used, its publication date and whether it was crossed through the euro. This is the auditable rate accounting and customs ask for, not a live market quote: it is published once a day and it is reproducible.

Flip one or more fair coins and get heads or tails, with the tally. Cryptographically random, not Math.random. For tie-breaks and A/B splits an agent has to decide on its own.

Roll dice with any number of faces and get every roll plus the total. Uses the platform CSPRNG with rejection sampling, so every face is equally likely - a plain modulo is biased. Standard dice notation: 3d6, 1d20.

Uniform random integers in any range you give, with or without repeats. Rejection sampling, so the ends of the range are not slightly less likely - which is what happens with modulo.

Generate strong random secrets from the platform CSPRNG, with the entropy in bits so you can judge them. Four alphabets, including one with no look-alike characters for anything a human has to read back.

Pick one or more items at random from a list you send, with or without repeats. The list is yours, so this is the unbiased chooser an agent needs when it has options and no reason to prefer any.

Shuffle a list into a uniformly random order with Fisher-Yates over the platform CSPRNG. Every permutation is equally likely, which sorting by a random key does not give you.

Everything the authoritative public sources say about one autonomous system number: the RDAP registry record via the ARIN-to-RIR referral chain (holder, status, allocation dates, abuse contact), the AS name, country and RIR from Team Cymru's DNS service, and the prefixes it is announcing right now in the global routing table, from RIPEstat, split into IPv4 and IPv6 with a sample. Keyless and free at the source. The natural second call after /v1/red/ip gives you an origin ASN.

Every currently valid TLS certificate publicly issued for one domain, read from the Certificate Transparency logs: issuer, validity window, days to expiry, revocation status and every hostname each certificate covers. It reports what was ISSUED, which is how you find certificates you did not know existed - not what the server is serving today. No API key, no account.

Checks one IPv4 address against nine public DNS blocklists at once — SpamCop, UCEPROTECT, PSBL, Barracuda, DroneBL and four more — and returns each list's raw return code plus the reason text the list itself publishes, not a score. Every list is re-checked on each call with its RFC 5782 test entry, so a dead list or one that refuses queries comes back as unavailable instead of reading as clean. Includes the delisting URL for each hit.

Everything DNS says about one domain in a single call: whether it is registered, its A and AAAA addresses, nameservers, MX servers, SPF and DMARC, DNSSEC, and whether it looks parked. Built for going through a list of domains one at a time. Live DNS, no API key, no account.

Scores one email address 0-100 with a verdict (ok, risky, disposable, undeliverable, invalid) and the evidence under it: every point lost is itemised with the field that proves it. 12,141 disposable domains from two maintained open lists pinned to a commit, 4,897 free providers, 1,018 role names, typo fix (gmial.com > gmail.com), live MX/SPF/DMARC. No SMTP probe, no API key.

Looks up one IPv4 or IPv6 address against the two authoritative, keyless public sources: RDAP registry data (who holds the block, in which country, since when, abuse contact) via the ARIN-to-RIR referral chain, and the origin ASN from Team Cymru's DNS service. No city-level geolocation and no VPN/proxy detection — those need a licensed commercial database this endpoint deliberately does not use. Built for going through a list of IPs one at a time.