01 / AI SEARCH API
Real-time retrieval built for LLMs
One call returns ranked snippets, citations and full-text Markdown — ready to drop into a context window.
- Citations
- Markdown
- Deep / fast modes
NEWMCP Server is now GA
The infrastructure your agents run on
AI search, web crawling, real-time market data and an agent runtime behind one API. You build — Cexa handles scaling, rate limits and source sync.
production · ap-east-1
AG
Your agent
claude · gpt · custom
Online879 rpm
MC
MCP Server
npx cexa-mcp
Online41 tools
GW
Cexa Gateway
api.cexa.ai/v1
Online12,526 r/s
SE
AI Search
POST /v1/search
Onlinep50 172ms
CR
Crawler fleet
POST /v1/crawl
Online2,224 pg/s
MK
Market stream
WS /v1/markets
Online64 venues
RT
Agent Runtime
POST /v1/agents/run
Online432 runs
Trusted by AI teams and quant desks
PLATFORM
Everything an agent needs underneath — from retrieval to execution
02 / CRAWLING
Any page, out as your schema
A distributed fleet handles JS rendering, anti-bot and retries, returning JSON that matches your schema.
- JS rendering
- Schema extraction
- Site crawl
03 / MARKET DATA
Multi-venue quotes, aligned to the millisecond
One REST and WebSocket interface across crypto and equities, with live streams and historical replay.
- REST + WebSocket
- Historical replay
- 60+ venues
04 / AGENT RUNTIME · MCP
A managed runtime, MCP in one line
State, memory, tool orchestration and sandboxed execution out of the box — every capability also exposed as MCP tools.
- Memory
- Sandbox
- MCP
- Search P50 latency
- <200ms
- Uptime SLA
- 99.99%
- Pages crawled daily
- 2B+
- Venues & sources
- 60+
QUICKSTART
Your first request in five minutes
Official Python and TypeScript SDKs — fully typed, with retries and streaming built in.
Read the docs →# pip install cexa
from cexa import Cexa
client = Cexa() # reads CEXA_API_KEY
res = client.search("HBM4 production status", depth="deep", max_results=5)
page = client.crawl(res.results[0].url, format="markdown")
quote = client.markets.quote("BTC-USDT")
print(res.results[0].title, quote.last)// npm install @cexa/sdk
import { Cexa } from "@cexa/sdk";
const cexa = new Cexa(); // reads CEXA_API_KEY
const res = await cexa.search("HBM4 production status", { depth: "deep" });
const page = await cexa.crawl(res.results[0].url, { format: "markdown" });
const quote = await cexa.markets.quote("BTC-USDT");
console.log(res.results[0].title, quote.last);curl https://api.cexa.ai/v1/search \
-H "Authorization: Bearer $CEXA_API_KEY" \
-H "Content-Type: application/json" \
-d '{ "query": "HBM4 production status", "depth": "deep" }'FAQ
Frequently asked questions
What is Cexa?
Cexa is infrastructure for AI agents. It puts AI search, web crawling, real-time market data and a managed agent runtime behind one API, so you build the agent and Cexa handles scaling, rate limits and source sync.
What does the AI Search API return?
A single call returns ranked, structured snippets with citations and full-text Markdown that you can drop straight into an LLM context window. Choose deep or fast mode depending on how much retrieval you need.
Can I use Cexa from Claude, Cursor or another MCP client?
Yes. Every Cexa capability is exposed as MCP tools for clients such as Claude and Cursor. Run npx cexa init to get started, and start the MCP server with npx cexa-mcp.
Which SDKs are available?
Official Python and TypeScript SDKs are available. They are fully typed, with retries and streaming built in. You can also call the REST API directly with cURL.
How does pricing work, and is there a free plan?
Pricing is usage-based: per 1,000 queries or pages, per agent-hour and per 1M market-data messages. The Developer plan is free with 1,000 calls a month and no credit card, and volume discounts apply automatically as you grow.
Do you offer dedicated clusters or private deployment?
Yes. The Enterprise plan includes a dedicated cluster or on-prem deployment, SSO and audit logs, a 99.99% SLA and a dedicated engineer. Talk to sales for details.