AnySearch
AnySearch is an API-first search infrastructure built for AI agents rather than people. A single endpoint routes each query to vertical sources — finance, code, security, academia and more — and returns structured, ready-to-use results through MCP, Skill or REST.
What is AnySearch?
AnySearch is a search infrastructure designed for AI agents rather than for people. It is published by Any Limited, which presents itself on its About page as an applied AI lab, and its stated premise is that agents do not need web pages — they need information that is reliable, structured and machine-readable, with a known provenance, a known freshness and a form they can act on directly.
The product's argument is that ordinary web search hits a wall. Stock prices are computed in real time, legal documents sit in paid databases, and code is scattered across millions of repositories; none of that is reachable by crawling public pages. AnySearch therefore goes to those sources directly and exposes them behind a single entry point. The domains it advertises are finance, academia, code, security, business, medical and health, patents, energy, environment, agriculture, travel and gaming, described as an expanding list.
Five capability blocks structure the offer: security and privacy by design, unified integration, agent-native output, smart intent routing, and full-spectrum coverage. In practice a query is classified by intent, routed hierarchically to the relevant sources, then merged and re-ranked across domains. Results are scored on authority, diversity and freshness before a cross-source fusion algorithm orders them. Output arrives as compact Markdown through MCP or the Skill, or as JSON with a constant schema through the REST API — a shape the vendor claims cuts token waste substantially.
There are three ways in. A remote MCP server needs one line of configuration. A Skill installs inside the agent and runs on Python, Node.js or a shell. A REST endpoint, POST /v1/search, is also available, alongside a deep-search variant. Four MCP tools are exposed: search, get_sub_domains, batch_search and extract. Anonymous access works without any key, at lower rate limits.
On privacy the vendor is unusually explicit: anonymous usage, no tracking, no telemetry, no logging, queries never stored persistently and never used for training. Benchmarks against Parallel and Brave are published on the homepage, though the vendor ran them itself.
What it does
- Query several vertical data sources through one call instead of one connector per source
- Route each query automatically to the right sources based on detected intent
- Return results already structured for an agent to consume, with no parsing step
- Extract the full content of any URL and receive it as Markdown
- Run up to five independent searches in parallel, where one failure does not block the others
- Plug the tool into an agent through MCP or a Skill with no development work
- Call the REST API anonymously, without an account or an API key
When to use AnySearch / When not to
A quick filter to help you decide if AnySearch is the right fit.
When to use AnySearch
- Developers building AI agents that need a single, reliable information layer instead of one connector per data source
- Engineering teams whose agents must reach beyond the open web — real-time market data, paid legal databases, code across millions of repositories
- Security engineers triaging alerts, who need threat intelligence cross-checked across several professional sources in one query
- Product managers and competitive-intelligence analysts running recurring market, funding and user-sentiment research
- Small teams and solo builders prototyping on a budget: 1,000 requests a day and 20 QPS are free, and the API answers without any sign-up
When not to use AnySearch
- Anyone looking for a consumer search engine: there is no public search interface, only an API and agent integrations
- Buyers who require a contractual GDPR commitment, a customer-facing DPA, an Article 27 representative or a named DPO — none of these exist here
- Procurement teams that gate vendors on SOC 2 or ISO 27001, since AnySearch claims no security certification at all
- Organisations that need a predictable paid price today: the Professional tier is still marked Coming Soon and Enterprise runs through sales
- Anyone under 18, or any project that needs to switch off the content safety filter, which the terms declare impossible to disable or bypass
How to use AnySearch
A typical end-to-end flow, from setup to results.
- Decide how your agent will connect: remote MCP server, installed Skill, or direct REST calls
- Try the service first without an account — anonymous access works at lower rate limits
- For MCP, add one configuration entry pointing at https://api.anysearch.com/mcp; Streamable HTTP is native, while SSE and stdio clients need a local proxy
- For the Skill, either hand your agent the installation prompt pointing at the published skill-install document, or download a release from the anysearch-ai GitHub repository
- Move the unpacked Skill into your agent's skills directory — the paths for Claude Code, OpenCode, Cursor and Windsurf are documented
- Check your runtime before running the Skill: Python 3.6 or later with requests, Node.js 12 or later, or PowerShell 5.1 and bash 3.2 with jq and curl
- Create a free API key from the console when you want higher limits, or register in a single call with a real email address that receives the password
- Supply the key as a CLI flag or Authorization header, an ANYSEARCH_API_KEY environment variable, or an .env file — in that order of priority
- Call get_sub_domains before any vertical search to retrieve the valid sub-domains and their parameter schemas
- Handle rate limits: a breach returns HTTP 429, and the vendor recommends exponential backoff
Pros & Cons
Pros
- One API replaces a separate connector for every vertical source, and the output needs no parsing
- You can test it immediately: anonymous access works with no account, and the free tier allows 1,000 requests a day at 20 QPS
- Native MCP and Skill integrations are documented for named clients including Claude Code, Claude Desktop, OpenCode, Cursor, Windsurf, VS Code Copilot and Cline
- The privacy position is explicit and detailed: queries are not persistently stored, never used for training and never sold
- Retention periods are published line by line with actual figures rather than vague commitments
- Benchmarks are published with their methodology — datasets, model and question count are all stated
- The Skill is open source under Apache 2.0 on GitHub, and a public status page reports availability
Cons
- The benchmarks are produced and published by the vendor, with an AI model as judge — they are not independent measurements
- No paid price is public: the Professional tier is marked Coming Soon and Enterprise requires contacting sales
- No security certification is claimed at all, neither SOC 2 nor ISO 27001, despite a heavily security-oriented pitch
- Corporate transparency is thin: no postal address, no registration number, and founders identified only by first name or handle
- The legal frame is awkward for a European buyer — Hong Kong law, HKIAC arbitration, and a liability cap of HKD 1,000 on the free tier
- Announced payment methods are limited to Alipay and WeChat Pay, and purchased quotas neither carry over nor get refunded
- The product is very young, cloud infrastructure subcontractors are named only by category, and the SLA referenced in the terms is not published on the site
Pricing & Plans
A permanent free plan is available at USD 0 per month, covering 1,000 requests per day, a 20 QPS rate limit per key, structured outputs and community support. Beyond that, no paid price point is published: the Professional tier, marketed as Search Pro, is announced as Coming Soon, and the Enterprise plan is only available through a sales conversation. The terms state that pricing is governed by the published pricing page and that any change is announced at least 30 days in advance. Announced payment methods are Alipay and WeChat Pay.
- structured outputs
- 1
- 000 requests/day
- 20 QPS rate limit per key
- community support
- higher-quality search results
- more API requests
- higher request limits
- deeper vertical search
- custom API usage limits
- enterprise-grade rate limits
- enterprise-grade security
- privacy and compliance
- advanced vertical search
- dedicated account manager
- priority technical support
Data, GDPR & hosting
A consolidated view of how AnySearch handles your data.
GDPR overview
The privacy policy is built on Hong Kong law — the Personal Data (Privacy) Ordinance — and names the Office of the Privacy Commissioner for Personal Data as the complaint authority. GDPR appears only as an additional section: EU and UK users are granted portability, erasure, restriction of processing and the right to object to automated decisions based on legitimate interests. Beyond that, the site never claims GDPR compliance in so many words, and never denies it either. No Article 27 representative is designated, no data protection officer is named, and no customer-facing data processing agreement is offered — the processing agreements mentioned bind the vendor to its own subcontractors. Rights requests are answered within 15 business days.
Who owns the data?
Under the terms, the search queries you submit through the API remain yours in intellectual property terms; you grant Any Limited only the limited licence needed to run the service. The platform itself — software, code, interface design, trademarks and documentation — belongs to the vendor. On the results side, AnySearch positions itself as a data processor acting on your instructions, while you are the data controller responsible for whatever you do next. Because results come from third-party sources, the vendor gives no warranty at all about who owns the intellectual property in them, and leaves that assessment entirely to you.
Reuse rights
You may reuse search results without asking permission, but the vendor shifts the whole legal burden onto you: results originate from third-party sources, and AnySearch warrants nothing about their accuracy, completeness, timeliness, legality or data-protection compliance. You must check that your own acquisition, storage and processing of results complies with the law that applies to you, including running legitimate-interest assessments or a DPIA where personal data is involved. The terms also draw hard commercial limits: you may not resell or sublicense API access without written consent, build a product that directly competes with AnySearch, or bulk-scrape results to assemble a rival dataset.
Data retention & training
Hosting summary
The vendor describes servers spread around the world and states that personal data may be processed and stored across several jurisdictions, without naming a country for the main service. Only two locations are named anywhere. Data collected by the vendor's own tracking SDK is stored within the Hong Kong Special Administrative Region and is declared not subject to cross-border transfer; should that change, the vendor undertakes to comply with the Personal Data (Privacy) Ordinance and to obtain explicit consent. Separately, anonymised browsing data sent to Google Analytics reaches Google servers in the United States, with IP anonymisation enabled, Google Signals disabled and a 14-month retention. The vendor states that transfers only happen through lawful mechanisms and that its protective measures apply uniformly wherever processing occurs. Note that the site's DNS resolves to an Amazon address geolocated in Paris; that is a CDN edge node and says nothing about where data actually lives.
Things to keep in mind
Risks and trade-offs to weigh before adopting AnySearch.
- The performance figures on the homepage are self-published and self-judged; treating them as independent measurement would be a mistake
- The vendor disclaims any warranty on the accuracy or legality of results and pushes the whole compliance burden onto you, including DPIAs where personal data appears
- Content safety filtering is mandatory and cannot be bypassed; the vendor admits legitimate results may be intercepted and declines liability for what goes missing
- Identifying the company is hard: no address, no registration number, founders listed by first name or handle only, and cloud subcontractors named by category alone
- Free-tier liability is capped at HKD 1,000, purchased quotas do not carry over between cycles, and paid plans are non-refundable in most circumstances
- An agent that treats structured, confident-looking results as settled fact will propagate errors faster than a human reading a list of links would
- The domain was registered in 1999 and later acquired, so the age of anysearch.com says nothing about the age of the company behind the product
Setup & Integrations
Technical difficulty
Low for a developer, out of reach for anyone else. The fastest route is a single line of MCP configuration, and the API answers anonymously with no account at all. Installing the Skill means downloading a GitHub release and moving it into the agent's skills directory, or handing the installation prompt to the agent. Python 3.6, Node.js 12 or a shell with jq and curl will do. The friction sits with SSE and stdio clients — Cursor, Windsurf, VS Code Copilot and Cline — which each need a local proxy.
Deployment
Integrations
Supported languages
Behind AnySearch
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement AnySearch.
Frequently asked questions
What exactly is AnySearch?
How is it different from a normal search engine?
Which domains does it cover?
How do I connect it to my agent?
Do I need an API key to start?
Are my search queries stored?
Will my queries be used to train AI models?
Which law applies, and is the service GDPR compliant?
Is there a minimum age?
Can the content safety filter be turned off?
Should you pick AnySearch?
AnySearch is young but already usable with very little friction. Anonymous access means you can call the API before creating anything, the free tier is generous at 1,000 requests a day, and three documented integration routes — MCP, Skill and REST — cover most agent stacks in use today. The promise is narrow and clearly stated: close the gap between what a crawler can index and the vertical sources an agent actually needs, from real-time market data to paid legal databases and code buried in repositories.
The published benchmarks are favourable, but they were run and scored by the vendor itself, with an AI model as judge. They are worth reading as a statement of intent rather than as evidence, and the sensible response is a trial on your own queries.
The weak point is corporate transparency. There is no postal address, no registration number, no full name for any of the four founders, no security certification and no public paid price. The legal frame sits in Hong Kong, with HKIAC arbitration, a liability cap of HKD 1,000 on the free tier, and a privacy policy built on the Personal Data (Privacy) Ordinance rather than the GDPR — European rights are granted as an add-on, without a designated representative or a customer-facing processing agreement.
None of this disqualifies the product for a developer wiring up an agent, where the cost of trying is close to zero and the technical proposition is genuinely distinctive. It does matter for a procurement process, where the missing certifications, the absent paid pricing and the unnamed infrastructure subcontractors are exactly the questions a reviewer will ask. Treat it as a promising piece of infrastructure to test now and to re-examine when the paid tier and the company details become public.
- Choosing a selection results in a full page refresh.
- Opens in a new window.