Exa

Exa is an AI-oriented search and retrieval platform with web, people, company, scholarly, and structured research data. It is aimed at people who want a practical way to move from an idea or business task to a repeatable result. The vendor pages describe capabilities including Neural and web search, Page contents and highlights, People and company indexes, Deep Search with citations, Agent search endpoint. Together, those features make the product useful beyond a one-off demonstration: a team can fit it into an existing process, hand work between people and AI, and keep the output connected to the systems where work actually happens.

The strongest use cases are Research agents, Company and contact enrichment, News monitoring, Supplying cited web context to applications. A sensible rollout starts with a bounded workflow whose inputs and expected output are easy to inspect. Teams should test accuracy on their own data, decide when a person must approve an action, and measure whether the tool saves completion time rather than merely generating more material to review. Technical buyers should also evaluate identity controls, auditability, retention, export options, and how usage limits behave during busy periods.

Pricing found on the vendor site was: Starter: Free; $20 signup credit plus $10/month; Developer: Pay as you go; Search $7/1,000 requests; Enterprise: Custom; Contents: $1/1,000 pages per content type; Deep Search: $12–$15/1,000 requests. These figures are snapshots from the September 2026 research run and can vary with annual billing, region, taxes, consumption, seats, or negotiated enterprise terms. Usage-based plans deserve a small production trial because agent loops, model calls, browser time, credits, or workflow executions may grow differently from ordinary seat-based software. Where a page did not expose a dependable figure in curl-fetched HTML, the profile says so and is flagged for manual verification.

MCP compatibility is a prominent part of the product: Exa includes MCP server access on its Starter plan and offers an official server for search and retrieval tools. This can reduce one-off integration work, but teams should still scope permissions and review every tool the agent can invoke. Overall, Exa is best evaluated against a real project using the listed capabilities, with a clear budget ceiling and success criteria. Its value is highest when the surrounding workflow, ownership, and review rules are defined before broad deployment.

Verified product detail

Exa is search infrastructure designed for AI applications rather than a consumer search box. It combines web ranking, page contents, highlights, answers, people and company indexes, and deeper cited research. The homepage reports Instant results under 180 milliseconds and offers vertical coverage for news, research, financial data, people, companies, and code. Its official MCP server exposes retrieval tools directly to compatible agents.

Key features to test

1. Neural, instant, automatic, and deep search modes. Verify behavior, limits, permissions, and plan entitlement with representative data and failure cases.
2. Contents, highlights, answers, and citations. Verify behavior, limits, permissions, and plan entitlement with representative data and failure cases.
3. People, company, research, news, and code indexes. Verify behavior, limits, permissions, and plan entitlement with representative data and failure cases.
4. Agent endpoint with effort controls. Verify behavior, limits, permissions, and plan entitlement with representative data and failure cases.
5. Official MCP search server. Verify behavior, limits, permissions, and plan entitlement with representative data and failure cases.

Current pricing and total cost

Search is $7 per 1,000 requests, Contents is $1 per 1,000 pages per content type, and Deep Search is $12–$15 per 1,000 requests. Agent pricing lists $0.10 per ACU, $0.005 per search tool call, and fixed effort from $0.012 Minimal to $1.00 X-high. Prices and entitlements can change; include taxes, overages, implementation, support, and review time in total cost.

Practical evaluation

Build a benchmark of 100 real queries covering fresh news, obscure documentation, known companies, and multi-source questions. Score top-ten recall, source authority, freshness, extraction quality, citation support, latency, and cost. Test deleted pages, duplicates, robots restrictions, and contradictory sources. Model the complete call graph because searches, content types, compute units, effort, and enrichment can all be billable. Email enrichment adds $0.02 and phone enrichment $0.07.

Honest strengths and limitations

• Purpose-built retrieval for agent applications
• Combines ranking, extraction, verticals, and research
• Published units support workload cost models

• Several billable dimensions complicate forecasting
• Coverage and freshness vary by niche
• Deep and enrichment workflows add cost and latency

• Supplying cited context to research agents
• Monitoring companies, people, and news
• Adding search and enrichment to applications

Alternatives and buying checklist

Compare Tavily, Brave Search API, Perplexity, LangChain with identical inputs, acceptance criteria, security constraints, and budgets. Confirm billing interval, cancellation, retention, export, training policy, regional hosting, uptime, and support. For agents that can write code or act in connected systems, start read-only, grant least privilege, require approval for consequential writes, log every action, and keep a tested rollback path. This review reflects vendor research performed September 14, 2026; select based on accepted outcomes after correction time and operating cost, not the most impressive first prompt.