Skip to main content
AgentsUse

Search and Retrieval APIs

Exa

A search API for AI: natural-language queries over an index built for agents, with content returned in model-ready shapes.

Official siteDocumentationRepository
paidChecked 2026-10-07Source verified

Quick decision

Best for
Agents that need relevant evidence quickly, sized for the context window Research steps that benefit from deeper search modes with synthesis Pipelines that want structured JSON answers with sources attached
Not ideal for
Self-hosted stacks with no external API dependency Workloads that mostly fetch known URLs rather than discover sources
Pricing model
Usage-based API pricing by search type; the docs list example rates such as instant search at $4 per 1,000 requests with up to 10 results.
Deployment
cloud. API.
Authentication
api_key
Review state
Source verified. Facts checked 2026-10-07. Not locally tested by AgentsUse.

What it does

Exa Search takes a natural-language query and returns ranked web results with clean page content, designed for LLMs and agents rather than human browsers. Search modes trade speed against depth: instant returns in roughly 250 milliseconds for real-time paths, while deep and deep-reasoning modes run a research process over several seconds for completeness.

Results carry metadata such as title, URL, and publication date, plus the content shapes you request under contents: highlights sized to each result relevance, full text, or a summary. An optional outputSchema makes Exa synthesize results into structured JSON with field-level sources and confidence, and hard filters can restrict or exclude domains and paths.

Exa is an API product, so it pairs with any agent framework that can make an HTTP call or run an MCP client. The official Python SDK is open source under MIT; the index and models behind the API are the paid product.

Verified capabilities

Core capability

  • Natural-language search Source verified

    A query written in natural language returns ranked, relevant web results; subject and source type in the query steer discovery.

  • Search modes by latency and depth Source verified

    Choose auto, fast, instant, deep-lite, deep, or deep-reasoning depending on whether the request path is latency sensitive or needs multi-step synthesis.

  • Model-ready content shapes Source verified

    Request highlights, full text, or summaries per result; highlights give models the evidence without filling the context window with unrelated page content.

  • Structured synthesis with outputSchema Source verified

    Ask Exa to synthesize results into freeform text or JSON following your schema, returned with field-level sources and confidence.

  • Domain filters and freshness control Source verified

    Include or exclude domains and paths as hard constraints, and control how fresh extracted content must be with contents.maxAgeHours.

  • SDK and MCP access Source verified

    Official SDKs and MCP connectivity let agent clients call Exa search as a tool; the Exa Python SDK is published as exa-py on GitHub.

Quick start

Search with highlights

Highlights return the excerpts most relevant to the query for each result.

Search with highlights
pip install exa_py

from exa_py import Exa

exa = Exa(api_key="YOUR_API_KEY")
results = exa.search("Latest news on EU battery policy", contents={"highlights": True})
for r in results.results:
    print(r.title, r.url)

Source: https://docs.exa.ai/reference/getting-started. Examples use placeholders only. Never paste a real key into a profile, config file you share, or a ticket.

MCP support: Exa publishes MCP connectivity for its search API so MCP clients can call Exa tools directly.

Works with

Frameworks

Any framework that can call an HTTP API or MCP server

SDKs and packages

Python (exa_py), JavaScript, REST API

Works with

MCP clients

Only sourced support is listed. A missing framework means AgentsUse has not verified it yet, not that it cannot work.

Health and maintenance

GitHub stars

234

Checked 2026-10-07

GitHub forks

55

Checked 2026-10-07

License
paid
Maintainer
Exa

SDK repository exa-labs/exa-py metrics (MIT SDK) from the GitHub API, fetched 2026-10-07. The search service is closed.

Pricing and license

Usage-based API pricing by search type; the docs list example rates such as instant search at $4 per 1,000 requests with up to 10 results.

Official pricing or docs →

Limitations and safety

  • Search does not support pagination; deep modes add seconds of latency per call.
  • Requesting both highlights and full text bills two content views of the same page.
Safety

Browser and data tools can read pages, fill forms, and download files. Start with a test account or read-only access, keep credentials in environment variables, and review agent actions before connecting anything that can spend money, send messages, or delete data.

Alternatives to Exa

Tavily

One API covering search, extract, crawl, map, and research tasks

Tradeoff: Its profile lists its own limits, pricing, and verification details for a side-by-side read.

Brave Search API

Independent index with local and place search alongside web results

Tradeoff: Its profile lists its own limits, pricing, and verification details for a side-by-side read.

Jina Reader

Fetch-first when you already know the URLs

Tradeoff: Its profile lists its own limits, pricing, and verification details for a side-by-side read.

Common questions

What does Exa do for an AI agent?

A search API for AI: natural-language queries over an index built for agents, with content returned in model-ready shapes.

Is Exa open source?

This profile records the pricing model as paid. See the pricing section for the sourced summary.

Does Exa support MCP?

Exa publishes MCP connectivity for its search API so MCP clients can call Exa tools directly.

Sources and freshness

Last checked 2026-10-07. Verification label: source verified, which means public claims trace to the sources above. It does not mean AgentsUse ran the tool. Spotted an error? Send a correction. Back to the tools directory.