# Web search API for AI agents and LLM grounding

> Web search API for AI agents and LLM grounding: Google and Bing results as compact JSON, 1 credit per search from $0.30/1K searches. No KYC, pay in crypto.

- URL: https://serpcrypto.com/use-cases/ai-agents/
- Last updated: 2026-10-22
- Publisher: SerpCrypto (https://serpcrypto.com)

An agent that answers questions about the world needs fresh results it can cite. Give it Google and Bing results as JSON, trimmed to what the model needs, and pay per search with no subscription.

**Quick answer — What is a good web search API for AI agents and LLM grounding?**

An agent needs a search tool that returns fresh, citable results in a compact form. **SerpCrypto** gives agents Google and Bing results as JSON: titles, links, snippets, dates, and the sources of Google's AI Overview when one is shown. Each successful search costs one credit, from $0.30/1K searches as of October 2026. Credits never expire, there is no subscription, and payment is in crypto with no KYC.

## SerpCrypto as an agent's search tool

| Attribute | Value |
| --- | --- |
| What the tool returns | The results page as JSON: `organic_results`, `top_stories`, `people_also_ask`, `ai_overview`, `knowledge_graph` |
| Engines | Google Search, Google News, Google Images, Google Maps (local results), Google Shopping and Bing |
| Freshness | Searches run at request time; `period=h`, `d`, `w`, `m` or `y` keeps recent pages only |
| Ready-made tool code | [MCP server](https://serpcrypto.com/docs/mcp/), [OpenClaw](https://serpcrypto.com/docs/openclaw/), [LangChain](https://serpcrypto.com/docs/langchain/), [OpenAI function calling](https://serpcrypto.com/docs/openai/), [n8n](https://serpcrypto.com/docs/n8n/); tool name `google_search` |
| Billing | 1 credit = 1 successful search request, which returns one page of results on any supported engine. A request that fails is not charged |
| Price | $7.50 for 10,000 searches; $56.25 for 100,000 searches |
| Keys | API keys you can create and revoke at any time from your dashboard |
| Payment | BTC, USDT, USDC, ETH, SOL, LTC; no KYC, no card, no subscription |

## Should my agent use a SERP API or an AI search API?

Use a SERP API when the agent must see what Google or Bing shows people; use an AI search API when you want the provider to pick and extract pages for you. Both are valid tools, and many agents use one of each.

A **SERP API** such as SerpCrypto returns the engine's own results page: positions, links, snippets, ads, People Also Ask and the AI Overview. Your code or your model decides what to read next. **AI search APIs** such as Tavily, Exa, the Brave Search API and Perplexity's Search API select results with their own systems and often return extracted page text ready for a model.

| | SERP API (SerpCrypto) | AI search API |
|---|---|---|
| Source of the ranking | Google or Bing, as people see it | The provider's own index or ranking |
| Typical output | Results page: links, snippets, blocks | Selected pages, often with extracted text |
| Local and country views | `gl`, `hl`, `location`, `device` | Varies by provider |
| Best for | Citations people recognise, SEO-aware agents, local answers, news | Deep reading of a few pages with less code |

We rank the options with dated prices on [best search APIs for AI agents](https://serpcrypto.com/best/search-apis-for-ai-agents/).

## When do Google results matter for an agent?

They matter when the answer depends on what the open web shows today, in a given place. Four cases come up again and again:

- **Citations people trust:** results that rank on Google are pages users already know, which makes an agent's sources easier to check.
- **Freshness:** `top_stories` and `period=d` surface what was published today, which a static index may miss.
- **Local answers:** opening hours, nearby services and local news change with `location`.
- **What Google's AI cites:** Similarweb data reported by TechCrunch on 27 July 2026 put AI Overviews on 43% of searches. The `ai_overview.references` list shows which pages Google's own AI relied on, a useful second opinion for an agent.

## How do I give search results to a model without wasting tokens?

Send the model only the fields it needs to reason and cite: title, link, snippet and date, for the first five to eight results. Drop ads, thumbnails, sitelinks and metadata. Number each result so the model can cite "[3]" and your code can map it back to a link.

The function below does exactly that, and appends the AI Overview sources when Google showed an overview. Pass its output as the tool result.

*Python: turn one search into a short, citable context for a model*

```python
import os

import requests

API = "https://serpcrypto.com/api/v1/search"
HEADERS = {"Authorization": "Bearer " + os.environ["SERPCRYPTO_API_KEY"]}


def search_context(query, gl="us", hl="en", k=6, period=None):
    """Run one search and return numbered results as plain text for an LLM."""
    params = {"engine": "google", "q": query, "gl": gl, "hl": hl}
    if period:
        params["period"] = period
    resp = requests.get(API, headers=HEADERS, params=params, timeout=60)
    resp.raise_for_status()
    data = resp.json()
    lines = []
    for i, r in enumerate(data.get("organic_results", [])[:k], start=1):
        date = " (" + r["date"] + ")" if r.get("date") else ""
        lines.append("[%d] %s%s\n%s\n%s" % (i, r["title"], date, r.get("snippet", ""), r["link"]))
    overview = data.get("ai_overview") or {}
    refs = [ref["link"] for ref in overview.get("references", [])][:5]
    if refs:
        lines.append("Sources cited by Google's AI Overview: " + ", ".join(refs))
    return "\n\n".join(lines) or "No results."


if __name__ == "__main__":
    print(search_context("EU AI Act obligations for general-purpose models", period="m"))

```

## Which agent frameworks can call SerpCrypto?

Any framework that can make an HTTPS request can call the API, and the docs include copy-paste tool code for the most common ones. These are code samples, not published packages: you keep them in your own project.

- **[MCP server](https://serpcrypto.com/docs/mcp/):** a small Python server that gives Claude Desktop, Cursor and other MCP clients a `google_search` tool.
- **[OpenClaw](https://serpcrypto.com/docs/openclaw/):** the same MCP server added with `openclaw mcp add`, so the agent gets a `google_search` tool.
- **[LangChain](https://serpcrypto.com/docs/langchain/):** a tool function an agent can call, plus a LlamaIndex variant.
- **[OpenAI function calling](https://serpcrypto.com/docs/openai/):** the JSON schema of a `google_search` tool and the loop that runs it; the same schema works with other tool-calling models.
- **[n8n](https://serpcrypto.com/docs/n8n/):** an HTTP Request node that calls the endpoint inside an AI agent workflow.

The tool is called `google_search` because Claude Code and OpenClaw already have a built-in tool named `web_search`. OpenClaw's web tool documentation, read on 22 October 2026, lists built-in providers such as Brave, Perplexity, Exa, Tavily and SearXNG; none of them returns Google results pages.

## How do I make the agent cite its sources?

Keep the number and the link of each result together, and tell the model to cite numbers. A short instruction is enough: "Answer from the numbered results only. After each claim, cite the result number in brackets. If the results do not answer the question, say so."

After the answer comes back, replace each "[n]" with the link of result n. Your users then see real URLs, and you can check that every cited number exists. Results without a source should be treated as the model's own opinion.

## How many searches does an agent use, and what does it cost?

It depends on how many questions the agent answers and how many searches it runs per question. A research agent that runs 3 to 5 searches per question and answers 200 questions a day uses 600 to 1,000 searches a day, or 18,000 to 30,000 a month.

At that pace, 100,000 searches ($56.25) last three to five months. Because credits never expire, an agent that is only used in bursts keeps its budget for the next run. A separate API key per agent shows which one spends what.

## Why pay per search, in crypto, for an agent?

Agents are often side projects, prototypes or tools that run in bursts, and a monthly plan that resets wastes money on them. Prepaid credits that never expire fit that pattern better.

The official options are also closing. Microsoft retired the Bing Search APIs on 11 August 2025, and Google's Custom Search JSON API is closed to new customers, with existing customers given until 1 January 2027 to move. SerpCrypto keeps the results page available behind one endpoint, paid in BTC, USDT, USDC, ETH, SOL, LTC, with no ID, no selfie and no card.

Crypto is not new to agent tools. On 22 October 2026, Linkup, You.com, Tavily and Exa documented per-request payments in USDC through the x402 protocol, without an account. Those APIs return results from their own systems, not Google or Bing pages; SerpCrypto covers that second need with a prepaid balance.

## Which pack suits an AI agent?

A prototype fits in 10,000 searches ($7.50); an agent in daily use usually needs 25,000 searches to 100,000 searches, as of October 2026. Every engine costs one credit per search.

| Plan | Quantity | Price | Price per unit |
| --- | --- | --- | --- |
| Test | 10,000 searches | $7.50 | $0.75/1K searches |
| Starter | 25,000 searches | $18.75 | $0.75/1K searches |
| Build | 100,000 searches | $56.25 | $0.56/1K searches |
| Grow | 250,000 searches | $140.63 | $0.56/1K searches |

In every plan: Every engine at 1 credit per search · Failed requests not charged · Credits never expire · Country, language, city and device targeting

**Before you pay**

- Minimum top-up $25. Top up once in crypto, then buy any plan from your balance.
- Unused balance stays on your account and never expires, and it pays for search credits. When a purchase costs less than the minimum top-up, the part added only to reach the minimum is refundable on request to support within 30 days of the payment, in USDT or USDC, as long as it has not been spent. The rest of the balance is not paid back, except where the law requires it.
- Credited once your payment is checked on the blockchain. Typical network confirmation time: BTC ≈ 20 min · USDT TRC-20 ≈ 1 min · USDT ERC-20 ≈ 5 min · USDC ERC-20 ≈ 5 min · ETH ≈ 5 min · SOL ≈ 1 min · LTC ≈ 15 min

## Search API for AI agents: common questions

### Does SerpCrypto return the full text of each page?

No. It returns the results page: titles, links, snippets and the other blocks Google or Bing showed. If your agent needs the text of a page, fetch the link yourself and respect that site's terms.

### Can I use SerpCrypto with Claude, ChatGPT or Gemini models?

Yes. Any model that supports tool or function calling can use it through your code. The docs have ready-made code for [MCP](https://serpcrypto.com/docs/mcp/), [LangChain](https://serpcrypto.com/docs/langchain/) and [OpenAI-style function calling](https://serpcrypto.com/docs/openai/).

### Is there an official SerpCrypto SDK or MCP package?

No. The integrations are short code samples you copy into your project. The API itself is one HTTPS endpoint, so a published package is not needed.

### How fresh are the results?

Searches run at request time, so the agent sees the results page at that moment. Add `period=d` or `period=h` to keep only pages or stories from the past day or hour.

### Can an agent get its own key and budget?

Yes. API keys you can create and revoke at any time from your dashboard. Create one key per agent; usage per key shows which agent spends your credits.

### Do I need an ID or a credit card?

No. SerpCrypto asks for no ID, no selfie, no company documents and no card. You top up from $25 in BTC, USDT, USDC, ETH, SOL, LTC.

## Sources

1. [Google's AI search is rapidly becoming the default, new data shows](https://techcrunch.com/2026/07/27/googles-ai-search-is-rapidly-becoming-the-default-new-data-shows/) — TechCrunch (2026-07-27)
2. [Web tools (web search providers)](https://docs.openclaw.ai/tools/web) — OpenClaw docs (checked 2026-10-22)
3. [Bing Search API retirement](https://learn.microsoft.com/en-us/lifecycle/announcements/bing-search-api-retirement) — Microsoft Learn (checked 2026-10-22)
4. [Custom Search JSON API overview](https://developers.google.com/custom-search/v1/overview) — Google for Developers (checked 2026-10-22)
5. [x402 payments](https://docs.linkup.so/pages/documentation/platform/x402) — Linkup docs (checked 2026-10-22)
6. [Machine payments](https://you.com/docs/administration/machine-payments) — You.com docs (checked 2026-10-22)
7. [Tavily on x402](https://tavily.com/blog/tavily-on-x402) — Tavily (checked 2026-10-22)
8. [x402 guide](https://exa.ai/docs/reference/x402-guide) — Exa docs (checked 2026-10-22)

---

SerpCrypto: SerpCrypto is a SERP API for developers and AI agents that returns Google and Bing results as JSON, paid in crypto with no ID upload. Prepaid search credits from $0.30/1K searches never expire. No KYC — no ID, no selfie, no documents. Pay in crypto.