Skip to content
Independent search for people and AI agents

Every search
adds a thread
to our own web.

Search the web, news, videos, images and code, hear what people say, and plug it all into your product or agent with one API.

  • Own crawler and index
  • No ads, no tracking
  • Filters that actually filter
  • MCP-ready
One API

Search, read and act
on the open web.

Everything an assistant, an agent or an app needs to find fresh information and turn pages into clean data. Pay per call, start free.

Search

Web, news, videos, images, code and people's voices, with time, language and site filters.

POST /v1/search

Scrape

Any public page as clean Markdown, without menus, ads or cookie banners.

POST /v1/scrape

Map & Crawl

Every URL of a site, or its pages as Markdown, depth and limits you choose.

POST /v1/map · /v1/crawl

Extract

Structured JSON from a page: prices, specs, contacts, following your own schema.

POST /v1/extract

Interact

A real browser that clicks, types, scrolls and waits, then returns the page.

POST /v1/interact

Monitor

Watch a page and know the moment its content changes.

POST /v1/monitor

MCP server

Give Claude, Cursor or any agent every tool above in one line of config.

POST /v1/mcp

Learns your users

A private index per account and per end user: your documents, what your users open.

POST /v1/index · /v1/feedback
Built for agents

Ten lines to a grounded answer.

Results come with dates, sources, thumbnails and an optional AI answer. Your agent gets facts it can cite, not a wall of links.

What people say

Not just what pages say.

Forums, open social networks and video sit beside the web. Ananke shows how many independent voices agree, so a copy-paste campaign does not pass for a consensus.

widely confirmed

Many independent voices, from more than one kind of source.

discussed

A few independent voices.

amplified

Many posts, few independent voices: read with care.

single voice

One source so far.

Pricing

Pay for what you weave.

One credit per search. Start with 500 free credits, no card.

All plans and top-ups
Playground

Try every endpoint, live.

Without a key you use the public sandbox (a few requests a day). With your key, calls spend your credits exactly like production.

Fill the form and run a request.

Documentation

AnankeSearch API

One REST API (and one MCP server) to search the open web by type, read any public page as Markdown or JSON, act in a real browser and watch pages for changes. Every endpoint takes and returns JSON over HTTPS.

Base URLFormatAuth
__BASE__JSON, UTF-8Authorization: Bearer as_live_…

Quickstart

1. Create a free account: you get 500 credits and your first API key. 2. Send a search. 3. Read results.

curl -X POST __BASE__/v1/search \
  -H "Authorization: Bearer as_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "best e-bike battery for winter", "max_results": 5}'

No key yet? Call the same endpoint without the header: the public sandbox answers a few requests per day so you can try it.

Authentication

Send your secret key in the Authorization header as a Bearer token. Keys start with as_live_, are shown once when created and are stored hashed: we cannot show them again. Create, name and revoke keys in your dashboard. Never put a key in front-end code; call AnankeSearch from your server.

Credits & limits

EndpointCredits
/v1/search (raw)1
/v1/search (mode answer)2
/v1/scrape, /v1/map, /v1/monitor1
/v1/crawl1 per page crawled
/v1/extract, /v1/interact2
/v1/index (add a document)1
/v1/feedback, DELETE /v1/indexfree

Every response returns credits_used and remaining_credits. Rate limits: 60 requests per minute on the free plan, 300 on paid plans. Top-up credits never expire.

Errors

StatusMeaningWhat to do
401Missing, malformed or revoked keyCheck the header and the key
402Not enough creditsTop up; detail.required says how many
403The URL is not a public addressOnly the public web can be fetched
422Invalid parameters, or a page that could not be readSee detail
429Rate or sandbox limit reachedRetry later, or use your own key

Errors are JSON: {"detail": {"error": "…"}}.

Search types

Each type returns only results of its kind. A video search returns videos (including video posts on social platforms), never encyclopedia pages padded in. When a plain query asks for a medium, for example elephant video or photos de Lomé, the type switches automatically and the response says so with "type_auto": true.

typeReturnsExtra fields
webPagesdate
newsArticles, freshest relevant firstsource, published, thumbnail
videosVideos from video platforms and social networksplatform, thumbnail, duration, author, published, embed
imagesImages with their pageimage, thumbnail, width, height, license, source
codeDocs, Q&A, repositorieslanguage, stars
socialWhat people say: forums, open networks, public postsplatform, author
curl -X POST __BASE__/v1/search -H "Authorization: Bearer as_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "elephant", "type": "videos", "max_results": 12}'

Filters

Filters combine with every type. They are applied, not suggested: with time_range: "week" nothing older than a week comes back, even if it would rank well.

{
  "query": "élections",
  "type": "news",
  "time_range": "week",
  "language": "fr",
  "site": "rfi.fr",
  "max_results": 10
}

Response

{
  "status": "success",
  "query": "elephant video",
  "type": "videos",
  "type_auto": true,
  "intent": "general",
  "lang": "en",
  "answer_direct": null,
  "results_count": 2,
  "results": [
    {
      "title": "…",
      "url": "https://www.youtube.com/watch?v=…",
      "snippet": "…",
      "kind": "video",
      "platform": "YouTube",
      "thumbnail": "https://i.ytimg.com/vi/…/hqdefault.jpg",
      "duration": "4:12",
      "published": "2026-11-10",
      "score": 0.812
    }
  ],
  "social": [],
  "echo": [],
  "credits_used": 1,
  "remaining_credits": 499,
  "latency_ms": 640
}
FieldDescription
results[]Ranked results. Always title, url, snippet, kind, score; plus the type's extra fields.
answer_directA short factual answer (a conversion, a definition) when one exists.
answerThe AI answer, in mode answer.
social[]People's posts: platform, url, title, text, author, date.
echo[]Claims people make, with how many independent voices agree: corroborated, discussed, amplified, single.
noticePresent when nothing of the requested type was found.
explain[]With explain: true: per result, the share of relevance, authority, source_quality, freshness and personal.

AI answers & full content

"mode": "answer" writes a short answer from the results, with numbered citations, in the question's language (2 credits). "include_content": true adds the Markdown of each result page, ready for your own model (RAG).

{"query": "how long does a lithium battery last in the cold", "mode": "answer", "max_results": 6}

People's voices

For opinions, experiences, troubleshooting and news, AnankeSearch adds what people say on forums, open social networks and public posts, and groups their claims by how many independent voices agree. Duplicated posts from different accounts count as one voice. It measures agreement, not truth: treat amplified with care. Use "social": "on" to always include voices, "off" to skip them, or "type": "social" to get voices only.

Personal index

Each account, and each end user you name with user, has a private index that nobody else can see.

CallWhat it does
POST /v1/feedback{"url", "query", "user"}: your user opened this result. Similar searches lift it. Free.
POST /v1/index{"url", "title", "text", "user"}: add your own page (help center, product sheet). It appears beside web results when it matches. 1 credit.
DELETE /v1/index{"user"} (optional): erase everything learned for the account or one user.
POST

/v1/scrape

A public page as clean Markdown. Parameters: url (required), max_chars (500–50 000, default 15 000). Returns title, content.

{"url": "https://example.com/article", "max_chars": 15000}
POST

/v1/map

Every internal URL of a site. Parameters: url, limit (1–500, default 100). Returns domain, total_links, links[].

POST

/v1/crawl

The pages of a site as Markdown. Parameters: url, limit (1–25 pages, default 5), max_depth (1–4, default 2). Returns base_url, pages_crawled, data[] with url, title, content. 1 credit per page.

POST

/v1/extract

Structured JSON from a page. Parameters: url, prompt (what to extract), optional schema (a JSON schema the output must follow). Returns data.

{
  "url": "https://shop.example/product/42",
  "prompt": "Product name, price with currency, and whether it is in stock",
  "schema": {"type": "object", "properties": {"name": {"type": "string"}, "price": {"type": "string"}, "in_stock": {"type": "boolean"}}}
}
POST

/v1/interact

Runs actions in a real browser, then returns the page as Markdown. Parameters: url, actions[], wait_ms (0–5000), max_chars. Actions: {"type": "click", "selector"}, {"type": "type", "selector", "text"}, {"type": "scroll", "direction": "down"}, {"type": "wait", "seconds"}.

{
  "url": "https://example.com/search",
  "actions": [
    {"type": "type", "selector": "input[name=q]", "text": "solar panels"},
    {"type": "click", "selector": "button[type=submit]"},
    {"type": "wait", "seconds": 1.5}
  ],
  "wait_ms": 1000
}
POST

/v1/monitor

Checks a page against its last known version. Parameter: url. Returns has_changed, change_count, last_checked. Call it on a schedule to watch prices, policies or job posts.

MCP server

AnankeSearch speaks the Model Context Protocol (JSON-RPC over HTTP) at /v1/mcp. Add it to Claude Desktop, Cursor or any MCP client; calls spend credits like the REST API.

{
  "mcpServers": {
    "anankesearch": {
      "url": "__BASE__/v1/mcp",
      "headers": {"Authorization": "Bearer as_live_YOUR_KEY"}
    }
  }
}
ToolDoes
anankesearch_web_searchSearch with type, time_range, language, site, mode
anankesearch_scrape_urlPage → Markdown
anankesearch_map_domainAll URLs of a site
anankesearch_crawl_siteSite pages → Markdown
anankesearch_extract_dataPage → JSON
anankesearch_interact_pageBrowser actions
anankesearch_monitor_pageChange detection
curl -X POST __BASE__/v1/mcp -H "Authorization: Bearer as_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"anankesearch_web_search","arguments":{"query":"elephant","type":"videos"}}}'

Recipes

Ground an assistant's answer (RAG)

Search with include_content: true, pass the Markdown of the top results to your model, cite the URLs.

A news watch for your users

type: "news", time_range: "day", language set to your user's. Deduplicate by url between runs.

A video finder

type: "videos" and show thumbnail, duration and platform; open url or play embed when present.

An image picker with licences

type: "images" and keep results whose license fits your use; always link back to url.

A support bot that knows your product

Add your help pages with POST /v1/index; they come first for matching questions, beside the web.

Price and policy watch

/v1/extract the fields you need, and /v1/monitor the page every hour; re-extract when has_changed.

Research what people really think

social: "on" and read echo: claims marked corroborated come from several independent voices.

Code samples

Pricing

Simple, per call.

One credit per search, 500 free credits to start. Monthly plans, or top-ups that never expire.

Monthly plans

Top-ups

Every plan includes
All search types and filtersweb, news, videos, images, code, voices
Scrape, map, crawl, extract, interact, monitorsame credits on every plan
MCP serverfor Claude, Cursor and any agent
Private index per account and per end usernever shared
Dashboard

Your account

Add credits
Credits left—
Requests—
Credits spent—
Average latency—

API keys

Your first call

curl -X POST __BASE__/v1/search \
  -H "Authorization: Bearer as_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "elephant", "type": "videos"}'
Open the playground

Recent requests

TimeEndpointQueryModeLatencyCredits
No requests yet.