Brave Search MCP Without an API Key: Get AI Mode Answers as JSON in 2026

AI answers are becoming the search result, and every engine writes its own. Brave Search is the interesting outlier: an independent index with a privacy-first audience, which means its AI Mode tells users a different story about your brand or topic than the mainstream engines do. I wanted that story as data, so I built the Brave AI Mode API on Apify. Give it a query and it returns the AI-generated answer, the sources it cites, and the supporting web results as one structured JSON record.

Disclosure: the Apify links in this post are affiliate links. If you run the Actor, I may earn a referral commission at no extra cost to you.

Doesn't Brave already sell a search API?

Yes, and credit where due: Brave's official Search API is one of the better ones an indie developer can get. But it's built around web results, it runs on API keys and plan quotas, and the AI Mode answer page, the thing users actually read, isn't what it hands you. If your question is "what does Brave's AI say about X, and who does it cite," you need the answer surface itself. This Actor scrapes exactly that and packages it like an API: query in, answer plus citations out, no Brave key to provision or babysit.

What the Brave AI Mode API returns

The Brave AI Mode API returns one record per query: the AI-generated answer text, the references it cites, and the supporting web results when present.

Field	Example	Notes
Query	what is quantum computing	The question you asked
AI answer	Quantum computing uses...	Full answer text as generated
Cited references	[{title, url}, ...]	The sources the answer leans on
Supporting web results	[{title, url, snippet}, ...]	The organic results beneath the answer
Country	us	Storefront of the search
Language	en	Interface language

One row holds the whole answer surface, which makes diffing runs over time straightforward.

Who this is for

Three audiences, in my experience. Marketing and SEO teams doing generative-engine monitoring, who need to know what AI answers claim about their brand and which domains get cited. Researchers and journalists who care specifically about the privacy-search ecosystem and its independent index. And agent builders who want an AI-grounded search tool without managing another vendor key.

The manual way, and where it breaks

You could check AI Mode by hand: type the query, read the answer, screenshot it, repeat next week. That survives about five keywords. Automating it yourself means a headless browser, because the answer streams in with JavaScript, plus session handling and markup that changes as Brave iterates on a young product. The iteration speed is the killer; new AI surfaces get redesigned monthly, and every redesign eats your selectors. I'd rather maintain that parser once, centrally, than have every reader maintain their own.

The faster way: run the Brave AI Mode API

Apify Console

1. Open the Brave AI Mode API and click Try for free.
2. Enter a query, or a queries list for batches, plus optional country and language.
3. Run it and download the answers as JSON or CSV.

REST

curl -X POST "https://api.apify.com/v2/acts/johnvc~brave-ai-mode-api/runs?token=YOUR_APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{ "query": "what is quantum computing", "country": "us", "language": "en" }'

Run and dataset endpoints are described in the Apify API docs.

Pull AI answers in Python

from apify_client import ApifyClient

client = ApifyClient("YOUR_APIFY_TOKEN")

run = client.actor("johnvc/brave-ai-mode-api").call(
    run_input={
        "queries": [
            "best password manager",
            "is vpn worth it",
        ],
        "country": "us",
    }
)

for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item.get("query"))
    print(item.get("answer"))

AI Mode answers show up for informational, question-style searches, so phrase your queries the way a user asking a question would.

A Brave Search API alternative for AI answers

The task Brave Search API alternative: AI answers as JSON frames the core trade: no key signup, no plan tiers, just per-query billing for the answer surface.

Check many keywords in one run

Check Brave AI answers for many keywords uses the queries array for batch monitoring, the shape you want for a keyword portfolio.

Track your brand in AI answers

Track brand mentions in Brave AI search answers is the generative-engine-optimization play: run your brand queries on a schedule and watch how the answer's framing shifts.

See who gets cited

Track sources cited in Brave AI answers inverts the question: which domains does Brave's AI trust for your topic, and is yours among them? Citation share is the new ranking, and this is how you measure it on Brave.

Spreadsheet route

Export Brave AI answers to CSV runs entirely in the Console for the no-code crowd.

Brave Search MCP, without an API key

This is the setup I use most. Registered through Apify's MCP server, the Actor becomes a tool that Claude, Claude Code, and Cursor can call, so your agent can consult Brave's AI Mode mid-conversation with no Brave key involved. The task Brave Search AI for Claude via MCP, no API key walks through the config, and there's more on Claude itself at claude.ai.

FAQ about the Brave AI Mode scraper

What does the Brave AI scraper cost per query?

A cent and a half per query resolved, covering the answer, its citations, and the supporting results in one record. There's no subscription, and the free credit on a new Apify account funds a decent first monitoring batch.

Why use this scraper instead of the official Brave Search API?

Different artifact. The official API serves web results under key-and-quota plans; this scraper captures the AI Mode answer surface, citations included, which is the thing GEO monitoring actually measures. Plenty of projects sensibly use both.

Can Claude Code call the Brave scraper as a tool?

Yes. Over MCP it works from Claude, Claude Code, and Cursor alike, and the no-API-key task above is the fastest path to a working setup.

How do I schedule this scraper for weekly answer tracking?

Save your keyword batch as a task, attach an Apify schedule, and each run appends a dated snapshot you can diff for changed answers or lost citations. Start from the Brave AI Mode API.

What are the limits of an AI answer scraper?

AI Mode doesn't fire on every query; navigational and transactional searches often return no answer, and that's data too. Answers are also generated fresh, so expect wording variation between runs even when the substance holds. Track claims and citations rather than exact strings.

More from Truffle Pig Data

The same monitoring pattern works on other engines' AI surfaces: the Google AI Overview API for Google's answer box, the Bing Copilot API for Microsoft's, and the Naver AI Overview API for Korea's leading engine.

Wrapping up

What Brave's AI says about your topic is already influencing users; the only question is whether you're reading it as data. The Brave AI Mode API makes that a one-query experiment.