# Hagglyai > Hagglyai tells AI agents which paid API to use for a job and what it really costs. It compares pay-per-call (x402) endpoints, Apify actors (payable over x402) and MCP servers on price per call, live uptime, response time and real paid usage. Free to query: no API key, no sign-up. Hagglyai is read-only. It never calls or pays a provider for you: use it to choose, then call the chosen endpoint yourself. ## Quick start 1. Ask for the best option for your job, with a budget per call: `GET https://hagglyai.com/api/v1/best?task=&max_price_usd=&format=md`, or the MCP tool `find_best`. 2. Read the **Recommended** line. It gives the endpoint as `METHOD URL`, its price per unit and its score. If the task matched no category, the answer lists the category ids: retry with one of them. 3. Before paying an endpoint you haven't used, check it: `GET https://hagglyai.com/api/v1/endpoints/lookup?url=` or the MCP tool `check_endpoint`. Look at its warnings and recent checks. 4. Call the endpoint. It answers `402 Payment Required` with its price, asset and payout wallet; pay with any x402 client and retry the request. Tip: always send a budget (`max_price_usd`) and hide warnings you can't accept, e.g. `hide_flags=wallet-mismatch,few-repeat-buyers`. Name the platform and the job in your task ("Instagram comments", "YouTube transcript", "Claude chat completion"): Hagglyai narrows to matching sub-categories (see below). Example tasks that match a category: "web search", "scrape a page to markdown", "BTC price", "ERC20 balance", "company lookup by domain", "weather in London", "crypto news", "chat completion". ## Ways to query ### MCP (recommended for agents) - Endpoint: `https://hagglyai.com/mcp`. Streamable HTTP, stateless, no auth. Install in Claude Code: `claude mcp add --transport http haggly https://hagglyai.com/mcp` - Claude on the web, desktop or mobile: add Hagglyai from Claude's connector directory: https://hagglyai.com/add/claude?via=llms - All tools are read-only and answer in markdown. - `find_best(task, max_price_usd?, hide_flags?, recent?)`: matches a plain-language task to a category, narrows to the platform and task it names (sub-categories), and returns the recommended endpoint plus up to 5 ranked options. - `compare_category(category, max_price_usd?, limit?, hide_flags?, recent?, platform?, task?, rail?)`: the ranked list for one category id, best value first. `limit` 1–50, default 15. `platform` and `task` take arrays of sub-category ids, `rail` an array of `x402` / `apify`. - `most_popular(category, max_price_usd?, hide_flags?, recent?, platform?, task?, rail?)`: the 3 endpoints in a category with the most paying wallets in 30 days, labelled Most popular and Popular. Endpoints carrying a warning are held back and named. Popular is not best value: check the score. - `check_endpoint(url)`: score, grade, uptime, the last 10 checks, listed price, paid usage and warnings for one endpoint URL (use the URL exactly as listed). - `list_categories()`: every category id with its endpoint count, cheapest price and sub-category ids (with seller counts). ### JSON API (OpenAPI: https://hagglyai.com/openapi.json) - `GET https://hagglyai.com/api/v1/best`: `task` (required, 2–300 characters), `max_price_usd`, `limit` (1–25, default 5), `hide_flags`, `recent`, `format=json|md`. Returns the matched `categoryId`, the `subcategories` it narrowed to and a `comparison`, or `categoryId: null` plus the list of categories. - `GET https://hagglyai.com/api/v1/categories`: all categories with counts, the cheapest price and each category's `subcategories` (ids, labels, seller counts). - `GET https://hagglyai.com/api/v1/categories/{id}`: `sort`, `limit` (1–100, default 25), `max_price_usd`, `hide_flags`, `recent`, `platform`, `task`, `rail` (each comma-separated). 404 for an unknown id. - `GET https://hagglyai.com/api/v1/endpoints/lookup?url=`: one endpoint with its score and last 10 checks. 404 if Hagglyai doesn't track it. ### Markdown - `https://hagglyai.com/index.md`: the categories table. - `https://hagglyai.com/compare/{id}.md`: one category, ranked. Takes the same filters as the category page (see below). - Send `Accept: text/markdown` to any page to get markdown instead of HTML. - `https://hagglyai.com/methodology.md`: how checks and scores work. - `https://hagglyai.com/x402/{question}.md`: one question answered from live data, e.g. `https://hagglyai.com/x402/cheapest-web-search-api.md` (the cheapest and the best x402 API per category; list at `https://hagglyai.com/x402.md`). ### Website (for people) - Search: `https://hagglyai.com/search?q=` redirects to the matching category page, or shows a "no provider sells that yet" page. - Category pages `https://hagglyai.com/compare/{id}` have sort tabs and filters; each endpoint has a page at `https://hagglyai.com/endpoint/{id}` with its checks and how to buy. ## Filters and sorting The same names work on pages, markdown, JSON and MCP. | Filter | Pages and markdown | JSON API | MCP | |---|---|---|---| | Sort order | `?sort=` | `sort=` | not available: tools use best value | | Maximum price per call (USD) | `max_price_usd=0.01` | `max_price_usd=0.01` | `max_price_usd: 0.01` | | Only endpoints checked in the last 7 days | `recent=1` | `recent=1` | `recent: true` | | Hide warnings | `hide=` (repeat for more) | `hide_flags=,` | `hide_flags: [""]` | | Sub-category: platform (any of) | `platform=` (repeat for more) | `platform=,` | `platform: [""]` | | Sub-category: task (any of) | `task=` (repeat for more) | `task=,` | `task: [""]` | | Payment rail (default both) | `rail=x402` or `rail=apify` | `rail=x402` | `rail: ["x402"]` | Sort names (the default is `best_value`): - `best_value` (Best value): highest score first, then most paying wallets, then lowest price. The Hagglyai Pick, if there is one, is listed first. - `cheapest` (Cheapest): lowest price per unit first (a call covering several units is divided by how many). - `most_reliable` (Most reliable): highest uptime over the last 7 days first (needs 3+ checks to count). - `fastest` (Fastest): lowest median response time first. Example: `https://hagglyai.com/compare/web-search.md?sort=cheapest&max_price_usd=0.005&recent=1&hide=wallet-mismatch` Prices are in US dollars, paid in USD stablecoin. Each category has a unit (one search query, one post or profile...): `unitPriceUsd` is the price per unit, `priceUsd` what one call costs. When a call covers several units, `unitsPerCharge` says how many and `unitBasis` how we know: `stated` (the listing says "up to N per call"; the unit price assumes a full batch), `example` (its example request asks for N items), `per-result` (Apify) or `call` (one call, one unit). Cheapest sorts and the Hagglyai Pick use the unit price; `max_price_usd` is a budget per call. "up to $X" means the seller charges at most $X. Testnet-only listings are never included. Rows with `rail: "apify"` are Apify actors: the price is per result, plus `startFeeUsd` once per run (and Apify platform usage when `platformUsageExtra` is true). Pay with an Apify prepaid token bought over x402 at agi.apify.com ($1 minimum), then POST the actor input to `url` with the token as a bearer token. Their `uptime7d` is Apify's reported 30-day run success, `probeCount7d` their run count and `payers30d` monthly users; Hagglyai does not probe them, so their score is capped at 90. ## Sub-categories Within a category, endpoints carry sub-category ids of two kinds: `platforms` (where the data comes from or what runs it: a social network, a model, a chain, a data source, a search engine, a marketplace) and `tasks` (what the endpoint does: comments, transcripts, forecasts, balances...). An endpoint can have several of each, or none. Ids are stable; labels may change. - Filter with `platform` and `task`. Several ids in one filter mean any of them; platform and task together mean both: `platform=instagram,tiktok&task=comments` is (Instagram or TikTok) and comments. - Only values that 3+ different sellers have are offered; unknown ids are ignored, not errors. - `find_best` does this for you: it finds platform and task words in your task text and narrows to them. If nothing matches, it drops the task, then the platform, and says what it used (`subcategories` in JSON, "Matched category: ... › Instagram › Comments" in markdown). - Ids are found from each listing's own wording, refreshed every 6 hours. A listing that doesn't name its platform or job carries none, so filtering can hide it: drop the filter if a list looks short. Example: `https://hagglyai.com/api/v1/categories/social-media?platform=instagram&task=comments&max_price_usd=0.01`, or MCP `compare_category({ category: "social-media", platform: ["instagram"], task: ["comments"] })`. Current ids per category (sellers in brackets; live list: `https://hagglyai.com/api/v1/categories` or `list_categories`): - `web-search`: search engine `google` (89), `exa` (35), `tavily` (23), `brave` (16), `bing` (13), `duckduckgo` (9), `perplexity` (8); task `results` (174), `answers` (93), `news` (76), `semantic` (29) - `web-extraction`: task `markdown` (416), `metadata` (218), `crawl` (167), `browser` (151), `robots` (115), `documents` (105), `structured` (100), `screenshots` (19) - `llm-inference`: model `openai` (69), `meta-llama` (42), `google` (40), `qwen` (37), `anthropic` (34), `deepseek` (25), `xai` (12), `mistral` (11); task `chat` (147), `reasoning` (28), `embeddings` (9), `images` (8), `speech` (7) - `crypto-prices`: coin `ethereum` (168), `bitcoin` (163), `solana` (130), `stablecoins` (52); task `spot` (246), `market` (160), `dex` (127), `ohlcv` (116), `derivatives` (42) - `onchain-data`: chain `base` (232), `ethereum` (118), `arbitrum` (70), `polygon` (70), `optimism` (62), `solana` (52), `bnb` (32), `bitcoin` (28); task `blocks` (219), `balances` (208), `transactions` (190), `tokens` (174), `rpc` (109), `risk` (46) - `enrichment`: task `company` (222), `people` (114), `registry` (106), `email` (80), `phone` (59), `domain` (30) - `weather`: data source `noaa` (74), `open-meteo` (49), `met-norway` (32); task `current` (190), `forecast` (155), `alerts` (72), `historical` (31), `air-quality` (27), `marine` (18) - `news`: task `headlines` (229), `crypto` (142), `topic-search` (99), `tech` (81), `sentiment` (65) - `social-media`: platform `youtube` (74), `x-twitter` (62), `instagram` (36), `tiktok` (35), `reddit` (29), `telegram` (20), `facebook` (18), `linkedin` (17), `bluesky` (16), `threads` (14), `pinterest` (5); task `posts` (134), `search` (74), `profiles` (57), `transcripts` (48), `comments` (40), `analytics` (31), `publish` (11) - `jobs`: source `ats` (22), `linkedin` (11), `indeed` (8); task `search` (59), `companies` (28), `salary` (22), `remote` (22) - `ecommerce`: marketplace `amazon` (31), `shopify` (18), `ebay` (11), `google-shopping` (10), `walmart` (9), `aliexpress` (4), `etsy` (3); task `products` (51), `prices` (33), `reviews` (33), `sellers` (32) - `maps-places`: source `google-maps` (38), `yelp` (6), `tripadvisor` (4), `openstreetmap` (3); task `places` (41), `geocoding` (32), `reviews` (29), `routing` (6) - `translation`: engine `llm` (12), `libretranslate` (5); task `text` (93), `detect` (44), `websites` (27), `documents` (18), `speech` (15), `subtitles` (11) - `messaging`: channel `email` (25), `sms` (12), `voice` (8); task `send` (32), `receive` (12), `bulk` (3) - `travel`: source `google` (10), `airbnb` (6), `booking` (3), `expedia` (3); task `airports` (42), `flights` (34), `status` (27), `hotels` (12), `reviews` (8), `ground` (4) - `stocks`: data source `sec-edgar` (174), `yahoo-finance` (19), `tradingview` (9), `alpha-vantage` (4), `finnhub` (3); task `forex` (267), `filings` (184), `quotes` (168), `history` (101), `fundamentals` (93), `earnings` (71), `screeners` (22), `options` (5) - `pdf-generation`: input `url` (24), `markdown` (13), `templates` (13), `html` (12), `images` (7), `office` (4); task `invoices` (14), `reports` (13), `screenshots` (6), `contracts` (4) - `captcha-solving`: captcha type `turnstile` (4), `recaptcha` (3), `hcaptcha` (3), `image` (3) - `sports-data`: sport `us-leagues` (27), `football` (25), `esports` (8), `combat` (7), `motorsport` (6), `golf` (6), `tennis` (5), `rugby` (5), `cricket` (4), `racing` (4); task `scores` (39), `odds` (27), `schedules` (25), `stats` (24), `standings` (22), `predictions` (21), `fantasy` (11) - `shopping`: task `checks` (39), `gift-cards` (31), `buying` (20), `returns` (6), `tracking` (6) - `speech-audio`: platform `openai` (40), `elevenlabs` (13), `deepgram` (9), `open-models` (9), `google` (7); task `transcription` (74), `text-to-speech` (69), `processing` (32), `captions` (21), `voices` (6) ## Warnings Results carry warning ids in `flags`. Warnings never hide anything by themselves; filter them out with the hide options above. - `wallet-mismatch` (Wallet mismatch): The payout wallet in its live payment answer differs from its catalogue listing on the same network. - `price-mismatch` (Price mismatch): The live price differs from the catalogue listing. - `few-repeat-buyers` (Few repeat buyers): Most payers called it only once, a pattern typical of crawlers or volume farming, so its usage counts half. - `unusual-seller-traffic` (Unusual seller traffic): This seller's paid usage looks manufactured: paid calls to trivial endpoints like ping or uuid, or near-identical payer counts on every endpoint. Its usage counts half. - `test-listing` (Test or demo listing): The name, host or route says this is a test or demo deployment. It can still be used, but it is never the Hagglyai Pick. - `resale` (Resold listing): This seller's prices are round prices plus a fixed markup (such as 15%): it relists other sellers' APIs. Buying from the original is usually cheaper, and its payers are not independent buyers, so its usage counts half. Informational flags (shown, never filterable): - `no-usage-data`: The catalogue reports no recent paid usage. - `testnet-only`: Only accepts testnet USD. - `reported-by-apify`: Scored from Apify's own run success rate and user counts; Hagglyai cannot check it without paying. ## Reading results - Every answer ends with its source: a `Source: hagglyai.com/compare/ · as of