Exa review: Search, contents and research API giving AI agents real-time web data

What Exa does

Exa is an API product for developers who build AI agents that need information from the web. The lineup has a Search API, a Contents API that returns full page text and token-efficient highlights, an Agent API for asynchronous deep research runs with structured outputs and citations, and a Monitors API that runs searches on a schedule and reports through webhooks. Exa Connect adds live data from third-party providers through one integration.

Search latency is configurable, starting with a lite mode at 450ms and going to deeper modes, so a team can trade speed against depth. The free plan also lists MCP server access, a Claude Connector and a ChatGPT plugin.

Pricing is usage-based, not a subscription. Enterprise is custom and adds items such as zero data retention and custom MSA and DPA terms.

Stated limits include 10 search queries per second on Starter, up to 25 on Developer, and 50 agent concurrency on both. Enterprise allows up to 1,000 results per search.

• Search API with configurable latency
• Full page contents for LLM context
• Token-efficient highlights
• Asynchronous deep research runs
• Structured outputs with citations
• Scheduled monitors with webhook updates
• Third-party data through Exa Connect
• MCP server access

Our take

Exa fits developers and teams building AI agents, coding assistants or enrichment workflows who want search and research billed by usage rather than by seat or subscription. The free Starter plan, with monthly credits and no payment method, lets you test before spending. It is a poor fit if you want a ready-made search app, because the pages describe an API product. Costs depend on the endpoint and the agent effort level, so budgeting takes some estimation. Higher limits, zero data retention and custom contracts sit with the Enterprise plan, which has no published price.