Pooled by design
Dozens of upstream keys behave like one big one. A Durable Object hands out slots, so your burst waits a few hundred milliseconds instead of failing with a per-key 429.
Power research agents
with the web and
the platforms behind it
Web search and page scraping on a pooled key ring, plus structured access to Reddit, YouTube, Amazon, Coupang, Naver, Instagram and app stores.
Ask a question, get evidence.
Ranked search results, page content and platform data from 14 sources, behind one key.
Search, scrape,
read the platforms
Every capability returns JSON or markdown, sits behind the same key, and shows up as an MCP tool. No HTML soup, no logins, no per-vendor SDKs.
headless browser pricing
Browserbase pricing and plansbrowserbase.com · 2026-08-30
Browserless: hosted Chromebrowserless.io · 2026-08-12
Web search
Ranked results with title, snippet, URL and date. Up to 20 results per query, batch up to 10 queries in one round trip.
GET anymorph.ai → markdown
# AI search optimization## How it works
Scrape
Any public URL to clean markdown, HTML or JSON. Links, images, CSS selectors and cache TTL pass through. Up to 10 URLs per call.
RedditYouTubeAmazonCoupangNaverInstagramApp StoreWalmart
Social & commerce
Posts, comments, reviews, products, trends and app listings from 14 platforms, through one social tool with per-platform actions.
Compare the top 3 headless browser services
Answer · 6 steps · 4 sources
Research agent
A ReAct loop over search, scrape and every platform source. Ask about a market, competitor or product and get an answer with the URLs it read.
Why anysearch
Most agents fail on the boring parts: rate limits, logins and pages that never render. Those parts are the product.
Platforms, not just pages
Korean and US commerce included. Coupang and Naver sit next to Amazon, Walmart, Sephora and Ulta, with reviews, products and search through the same tool.
Evidence, not vibes
The agent cites. It searches, scrapes and queries platforms in a loop, then summarizes with the URLs it actually read. Partial failures are reported, never hidden.
Quickstart
One key, one header
Sign in with Google, copy the key from the dashboard, send it as X-API-Key. The same key is your MCP bearer token for Claude Code and Codex.
Sign inCopy keyCall the API
# Search curl "https://anysearch.site/search?query=web+automation&max_results=5" \ -H "X-API-Key: $ANYSEARCH_API_KEY" # Scrape → markdown curl -X POST https://anysearch.site/fetch \ -H "X-API-Key: $ANYSEARCH_API_KEY" -H "Content-Type: application/json" \ -d '{"urls":["https://anymorph.ai"],"links":true}'
import requests r = requests.get("https://anysearch.site/search", params={"query": "web automation", "max_results": 5}, headers={"X-API-Key": ANYSEARCH_API_KEY}) print(r.json())
const r = await fetch("https://anysearch.site/fetch", { method: "POST", headers: { "X-API-Key": process.env.ANYSEARCH_API_KEY, "Content-Type": "application/json" }, body: JSON.stringify({ urls: ["https://anymorph.ai"], format: "markdown" }), }); console.log(await r.json());
# Claude Code claude mcp add anysearch --transport http https://anysearch.site/mcp \ --header "Authorization: Bearer $ANYSEARCH_API_KEY" # Codex (~/.codex/config.toml) [mcp_servers.anysearch] url = "https://anysearch.site/mcp" bearer_token_env_var = "ANYSEARCH_API_KEY"
Plans
Free while we're in preview
Sign in with Google and start calling the API in under a minute.
Free
$0/ month
Any Google account. 500 credits a month across search, scrape, social and the agent.
- 500 credits / month
- Search, scrape, agent, social
- MCP endpoint
- Up to 10 API keys
WorkspaceUnlimited
No credit limit
For @opactor.com and @anymorph.ai Google accounts. Everything in Free, without the monthly cap.
- No credit limit
- Everything in Free
- Shared pooled throughput
- Same keys, same MCP endpoint
1 credit = 1 search · 1 scraped URL · 1 social call. An agent run is 10. Failed operations are refunded.
FAQ
Frequently asked questions
The questions we get most often before the first API call.
How are credits counted?
A search or social call costs 1 credit. A scrape costs 1 credit per successfully fetched URL, 2 with a summary. An agent run costs 10. Credits are reserved up front and refunded for failed operations, including the failed URLs of a partially successful scrape.
What happens when an upstream provider rate-limits?
Upstream keys are pooled and scheduled by a Durable Object. A burst waits for a free slot instead of failing with a per-key 429. If the whole pool is cooling down you get an explicit error, not a silent retry.
Can I use it from Claude Code or Codex?
Yes. /mcp is a stateless Streamable HTTP MCP server. Add it with your API key as the bearer token and search, scrape, social and agent appear as tools. Long runs send progress heartbeats every 10 seconds.
What if a page only renders with JavaScript?
Empty or JS-only pages come back as an explicit quality error instead of blank content, and you are not charged for them. Pass fallback: "firecrawl" to retry those pages in markdown mode. Fallback results are labeled so you know where the content came from.
What are the request limits?
Search returns 1 to 20 results per query and batches up to 10 queries. Scrape accepts 1 to 10 public HTTP(S) URLs per call. An agent run has a 4-minute deadline and a budget of 30 URLs and 40 tool calls. Invalid inputs fail before anything is charged.
Which platforms are covered?
YouTube, Reddit, Google Trends, Instagram, Threads and TikTok on the social side. Amazon, Coupang, Naver, Walmart, Sephora and Ulta for commerce. App Store and Google Play for apps. Each platform exposes its own actions, and unknown actions fail instead of falling back to a generic search.