Exa: The Search Engine Built for AI Agents, Explained
Exa is a search engine built for AI rather than people. Instead of returning a page of blue links, it gives AI agents and apps the web pages, excerpts and structured data they need, through an API. Unlike many search APIs that wrap Google or Bing, Exa crawls and indexes the web itself and trains its own search models, which it says lets it compete on quality, speed and cost at once. Cursor, Cognition, HubSpot and OpenRouter use it, and it says more than 900,000 developers build on it. In May 2026, Exa raised $250 million at a $2.2 billion valuation.
The quick answer: Exa is one of the leading search APIs for AI agents, and the one most committed to owning its own index. It’s a strong default for developers building coding agents, research tools, sales prospecting and anything else where an AI needs current, relevant information from the web. It has a generous free tier, a hosted MCP server and connectors for Claude and ChatGPT. Its benchmark claims are its own, so test it on your queries. Everything below comes from Exa’s site, pricing and developer documentation, its Y Combinator profile and press coverage.
Exa at a glance
Checked against Exa’s site, pricing page, developer summary and Y Combinator profile on November 13, 2026.
| Exa | |
|---|---|
| What it is | A web search engine and data platform built for AI agents, sold as APIs |
| Products | Search, Contents, Agent, Monitors, Snapshot, Library and Connect |
| Index | Its own crawl of the web, plus structured data on companies, people, papers, filings and more |
| Works with | MCP clients like Claude, Cursor and Codex; a Claude connector and ChatGPT plugin; 50+ integrations |
| Free tier | $10 in credits a month, plus a one-time $10 bonus; no card required |
| Customers | Cursor, Cognition, HubSpot, OpenRouter, Monday.com |
| Company | Exa Labs, San Francisco; founded 2021 by Will Bryk (CEO) and Jeffrey Wang; Y Combinator S21 |
What Exa does
Exa’s about page calls the company “an applied AI lab building a perfect search engine.” Its argument is that AI agents will soon run vastly more searches than people do, and that they need search tuned for them: fast enough to call several times within one task, returning clean text rather than web pages full of ads and navigation, and able to understand what a query means rather than just matching keywords.
Exa’s developer summary describes it as “a real-time knowledge index and retrieval platform built for AI agents” that “builds and maintains its own index and search models.” Beyond the open web, Exa indexes structured collections it calls Exa Library, covering people, companies, places, job postings, earnings calls, SEC filings, stock and market data, research papers, patents, court opinions, products, events and more.
The products
| Product | What it does |
|---|---|
| Search | Finds relevant pages and returns text and highlights, with modes from low-latency Lite to multi-step Deep Search |
| Contents | Extracts clean, AI-ready text from any URL, including JavaScript-heavy pages and PDFs |
| Agent | Runs multi-step research, list-building and enrichment tasks with structured, cited results |
| Monitors | Runs searches on a schedule and sends new results by webhook |
| Snapshot | Searches the web as it was on a past date, for testing and backtesting |
| Connect | Pulls in data from third-party providers like Similarweb, Crunchbase and ZoomInfo through one integration |
A notable feature is highlights: instead of handing a model a whole web page, Exa uses a model trained to pick out the most relevant excerpts, which saves tokens. Exa says up to 25 trillion tokens a week passed to models are these excerpts.
Who uses it, and for what
Exa’s homepage quotes Cognition, the maker of the Devin coding agent: “Traditional search engine solutions weren’t enough for us. Exa powers all parts of Devin.” Its pricing page says Cursor uses Exa to fetch current documentation for more accurate code. Other named customers include HubSpot, which Exa says it powers an agentic lead engine for, plus OpenRouter and Monday.com.
Exa’s listed use cases span coding agents that need current docs and code, AI assistants that need live web context, sales and recruiting teams building lists of companies and people, financial and legal research with cited sources, and companies training AI models.
How developers connect
Developers use Exa through an API key and official Python and JavaScript SDKs. For AI tools, Exa runs a hosted MCP server that connects search, code search and research to Claude, Cursor, Codex, VS Code and other clients, which can be tried at limited usage without a key. It also has a Claude connector, a ChatGPT plugin, integrations with frameworks like LangChain, CrewAI, LlamaIndex and the Vercel AI SDK, and automation tools like n8n. Coding agents can even create an Exa account and API key on their own through Stripe Projects.
Pricing
Exa’s pricing starts with a free Starter plan: $10 in credits every month plus a $10 onboarding bonus, access to every endpoint, and no payment method required. After that, it’s pay as you go, with custom Enterprise plans for large customers.
| Endpoint | Base price |
|---|---|
| Lite Search | $4 per 1,000 requests, up to 10 results |
| Auto Search | $7 per 1,000 requests |
| Deep Search | $12 per 1,000 requests |
| Contents | $1 per 1,000 pages |
| Answer | $5 per 1,000 requests |
| Agent | $0.012 to $1.00 per run, depending on effort level |
Enterprise adds higher rate limits, single sign-on, SLAs, custom indexes, HIPAA with a BAA, and zero data retention, which requires a $12,000 annual commitment. Exa’s homepage says queries and results are never stored or trained on, while the pricing page lists zero data retention as an Enterprise feature, so teams with strict data rules should confirm which terms apply to them.
Benchmarks
Exa says its latest search version, Exa 3.0, leads on BrowseComp, DSQA and WideSearch, standard benchmarks for search agents, and is the best on its tests for code, finance and finding companies and people. These are Exa’s own results, and other search APIs publish benchmarks of their own. The practical approach is to run a sample of your real queries through two or three providers before choosing.
The company
Exa Labs was founded in 2021 by Will Bryk, its CEO, and Jeffrey Wang, and went through Y Combinator’s Summer 2021 batch. It’s based in San Francisco.
Exa raised an $85 million Series B led by Benchmark in September 2025, at a $700 million valuation, with Nvidia, Y Combinator and Lightspeed participating. In May 2026, SiliconANGLE reported a further $250 million at a $2.2 billion valuation, to buy infrastructure, train models and handle hundreds of thousands of searches per second. Bryk’s pitch for the company: “Most other search providers actually wrap other search engines and therefore cannot compete on quality/latency/cost.” Exa says it runs its own GPU cluster for training and is “on track to exceed Google-scale traffic and index size in 2027.”
How Exa compares
Exa’s main competitors are other search APIs for AI: Tavily, Parallel, the Brave Search API, and scraping-first tools like Firecrawl. Exa’s distinguishing traits are its own index and models rather than a wrapper, its structured data collections on people and companies, and its range from sub-second search to multi-step research agents. Brave also runs an independent index; Tavily and Parallel focus heavily on agent-ready results and research; Firecrawl is strongest at turning specific websites into clean data.
If you’d rather have an assistant that does the research for you, Carly searches the web as part of its work and connects to thousands of apps, so you can ask it from ChatGPT, Claude or a text message to look into a company before a meeting and save what it finds to your CRM.
Who Exa suits
Exa is a good fit if:
- You’re building a coding agent, research tool or AI assistant that needs fresh web information.
- You need to find and enrich lists of companies or people.
- You want search that understands meaning, with token-efficient excerpts.
- You want an MCP server or Claude connector you can set up in minutes.
Look elsewhere if:
- You only need to scrape specific known websites; a crawler like Firecrawl may fit better.
- You need guaranteed zero data retention without an Enterprise contract.
- You want a finished assistant rather than an API.
FAQ
What is Exa?
Exa is a search engine built for AI agents. It crawls and indexes the web itself, trains its own search models and gives developers APIs for search, page contents, research and monitoring.
Is Exa free?
Exa’s Starter plan gives $10 in free credits every month plus a $10 onboarding bonus, with access to all endpoints and no card required. Beyond that, pricing is pay as you go.
Does Exa have an MCP server?
Yes. Exa runs a hosted MCP server that connects its search and research tools to Claude, Cursor, Codex, VS Code and other MCP clients. It also has a Claude connector and a ChatGPT plugin.
Does Exa use Google?
No. Exa says it builds and maintains its own web index and search models rather than wrapping another search engine.
Who founded Exa?
Will Bryk and Jeffrey Wang founded Exa in 2021. It went through Y Combinator’s Summer 2021 batch and raised $250 million at a $2.2 billion valuation in May 2026.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
