Tavily vs Exa vs Firecrawl comes down to the job. Pick Tavily for the quickest general web search with ready to use snippets and simple credit pricing (1,000 free credits a month). Pick Exa when your agent must find specific kinds of pages, such as companies, people or research papers, by meaning; it also has the cheapest published per search rate here, $4 per 1,000 Lite searches, plus $10 of free credit every month. Pick Firecrawl when you need the full content of pages or whole sites as clean markdown, or want to self host.

Many agents pair a search API with Firecrawl: one finds pages, the other reads them. All three have Python and JavaScript SDKs and hosted MCP servers. Prices and limits below come from each vendor's official pages, checked on November 18, 2026, and the code matches tavily-python 0.8.5, exa-py 2.28.0 and firecrawl-py 4.57.0.

What's new (November 18, 2026): Exa now publishes a free tier and per request rates, so this guide compares all three on cost. Firecrawl lists paid plans from $19 a month and no longer lists the Question and Highlight formats we cited on November 9. Tavily's prices are unchanged.

Tavily vs Exa vs Firecrawl at a glance

ItemTavilyExaFirecrawl
Core jobSearch with answer ready resultsAI search engine, by meaningScrape, crawl and extract pages
Other endpointsExtract, map, crawl, researchContents, Answer, Deep Search, Agent, MonitorsSearch, map, interact, monitor, agent
Free usage1,000 credits a month$10 credit a month1,000 credits a month
Cheapest paid route$0.008 a credit$4 per 1,000 Lite searchesHobby, $19 for 5,000 credits
Open sourceNoNoCore AGPL-3.0, SDKs MIT
Hosted MCP with no accountYes, cappedYesAsked for a key

Tavily: answer ready search for agents

Tavily is designed as the search step in an agent loop. One call returns ranked results with relevant content, and you can ask for a short answer, full page markdown, or news and finance topics only. Other endpoints extract URLs, map and crawl sites, and run multistep research.

from tavily import TavilyClient

client = TavilyClient(api_key="tvly-YOUR_KEY")
response = client.search(
    "Model Context Protocol latest specification changes",
    search_depth="advanced",
    max_results=5,
)
for result in response["results"]:
    print(result["title"], result["url"])

Per Tavily's pricing page and credits documentation, the free plan gives 1,000 credits a month with no card, reset on the first of the month. A basic search costs 1 credit and an advanced search 2. Extract costs 1 credit per 5 successful URLs (2 at advanced depth), map costs 1 credit per 10 pages, and a Research request costs 4 to 110 credits on mini or 15 to 250 on pro. Pay as you go credits are $0.008. Monthly plans run from Project ($30 for 4,000 credits) to Growth ($500 for 100,000, or $0.005 a credit).

See whether the Tavily API is free for keyless calls and the student plan, and how to use Tavily in LangChain for the official tool package. Tavily's remote MCP server is https://mcp.tavily.com/mcp/; in Claude Code, run claude mcp add tavily-remote-mcp --transport http https://mcp.tavily.com/mcp/. Our guide to whether Tavily MCP is free covers the keyless header and OAuth.

Best for: teams adding web search to an agent for the first time. Weak spot: less control over results than Exa, and heavy crawls suit a dedicated scraper better.

Exa: search that understands meaning

Exa runs its own search engine, built to be queried by AI. You describe the page you want and Exa matches the meaning. The search API reference lists categories for company, people, publication, news, personal site and financial report, plus domain and date filters, with text, highlights or summaries in the same call. The lite type suits chat and voice, auto (the default) suits most work, and deep handles multistep questions.

from exa_py import Exa

exa = Exa(api_key="YOUR_EXA_KEY")
response = exa.search(
    "open-source sandboxes for running AI agent code",
    num_results=5,
    contents={"text": True},
)
for result in response.results:
    print(result.title, result.url)

Install with pip install exa-py or npm install exa-js. The hosted MCP server at https://mcp.exa.ai/mcp answered our test with no key; our roundup of free MCP servers for web search compares it with other no cost routes.

Exa's pricing page says it is pay as you go with no subscription. The free tier is $10 of credit (up to 2,500 Lite searches), reset to $10 on the first of each month, no payment method needed. Search costs $4 per 1,000 on lite and $7 on auto, covering up to 10 results; each extra result and each AI page summary adds $1 per 1,000. Deep Search costs $12 to $15 per 1,000, Contents $1 per 1,000 pages per content type, and Answer $5 per 1,000.

Best for: research, recruiting and sales agents that hunt for companies, people or papers. Weak spot: more options to learn, and summaries or deep types change the per call price.

Firecrawl: turn any site into clean data

Firecrawl solves the reading side. Give it a URL and it returns clean markdown, handling JavaScript, PDFs and common anti bot measures. Crawl follows links across a site, map lists URLs, extract pulls JSON with a schema, and Interact drives a real browser.

from firecrawl import Firecrawl

app = Firecrawl(api_key="fc-YOUR_KEY")
doc = app.scrape("https://modelcontextprotocol.io", formats=["markdown"])
print(doc.markdown[:500])

results = app.search("agent sandbox comparison", limit=5)

The core is AGPL-3.0 and the SDKs are MIT, per its GitHub repository, so you can self host the engine; the cloud adds managed proxies. On the pricing page, scrape and crawl cost 1 credit per page, map 1 credit per call, search 2 credits per 10 results, and Interact 2 credits per browser minute. JSON output or the prompt injection check adds 4 credits per page. A page that returns a 403 or 404 still costs 1 credit. The free plan has 1,000 credits a month and 10 scrape, map or search requests a minute. Paid plans run Hobby ($19 for 5,000 credits), Standard ($99 for 100,000), Growth ($399 for 500,000) and Scale ($749 for 1,000,000). Setup for coding agents is in how to use Firecrawl MCP.

Best for: RAG pipelines and agents that need full pages or whole sites, and teams that want to self host. Weak spot: search is secondary, and the AGPL matters if you offer a modified core as a service.

What 1,000 searches cost on each

Here is the arithmetic for 1,000 plain searches of up to 10 results, from the November 18 price lists, before extra content options.

Route1,000 searchesNote
Exa lite$4Text or highlights in the same call
Exa auto$7Default type
Tavily basic, pay as you go$8Advanced depth doubles it
Tavily basic, Growth plan$5Needs the $500 plan
Firecrawl, Hobby top up$102,000 credits at 1,000 per $5; scraping extra
Firecrawl, Standard top up$52,000 credits per $5 on the $99 plan

Free usage alone covers up to 2,500 Exa Lite searches, 1,000 Tavily basic searches, or 500 Firecrawl searches a month. Recheck the pricing pages before you set a budget.

What our keyless test showed

On November 18 we sent one query to each hosted MCP server with no account. Exa answered in 0.3 seconds with the official MCP blog post first. Tavily answered in 1.4 seconds, but its top hit was a third party mirror calling an older spec version the latest, so check what your agent cites. Firecrawl asked for an API key. One query is not a quality test; try a sample of your own.

Using them together

A common pattern is search, then read: Tavily or Exa finds the top URLs, and Firecrawl fetches the one or two that matter. Tavily's and Exa's content options often make the second step unnecessary, so measure first.

In an agent framework, wrap each SDK call as a function tool; our LangGraph vs CrewAI vs OpenAI Agents SDK comparison shows how. In coding agents, connect the hosted server over the Model Context Protocol (new to it? read what MCP is). For caching or merging two providers, build your own with our Python MCP server guide.

How to choose a web search API for AI agents

  1. General web search with minimal setup? Start with Tavily.
  2. Specific kinds of pages, found by meaning? Choose Exa.
  3. Full content from known URLs or whole sites? Choose Firecrawl.
  4. Must self host? Only Firecrawl's core runs on your servers; see self hosted Tavily alternatives for more.
  5. Lowest cost? Exa Lite has the lowest published rate and the largest free allowance. Cache repeated queries either way.

Related: E2B vs Daytona vs Modal for where agents run code, AI agent memory for what they remember, and Browser Use vs Stagehand vs Playwright MCP for clicking through pages.

From Ahzy Capital (our own products, labeled as such): Get the free AI Agent Cheat Sheet: 24 copy and paste AI agent setups for 8 types of small business, with prompts and workflows. We make and sell these, so they aren't independent picks.

FAQ

What is the difference between Tavily and Exa?

Tavily is a search API tuned to return answer ready content, with simple credits and 1,000 free a month. Exa runs its own engine that matches by meaning, with category filters for companies, people, papers and news. Tavily is simpler to start; Exa gives more control.

Is Exa free?

Exa has a free tier: $10 of credit at sign up, reset to $10 on the first of every month, no payment method required. Exa says that covers up to 2,500 Lite searches. After that it is pay as you go from $4 per 1,000 Lite searches.

Is Firecrawl a search engine?

Not primarily. Firecrawl is a scraping and crawling service that turns pages into markdown or structured data. Its search endpoint costs 2 credits per 10 results, but reading pages is its strength.

Which web search API is cheapest for AI agents?

At November 18, 2026 list prices, Exa Lite is cheapest per plain search at $4 per 1,000, against $8 for Tavily basic on pay as you go and $10 for Firecrawl at the Hobby top up rate. Larger Tavily and Firecrawl plans bring theirs to about $5.

Do Tavily, Exa and Firecrawl have MCP servers?

Yes: mcp.tavily.com, mcp.exa.ai and mcp.firecrawl.dev. In our test Exa worked with no key, Tavily worked with its keyless header, and Firecrawl asked for an API key.

Can I self host a web search API for agents?

You can self host Firecrawl's open source core for scraping, crawling and extraction. Tavily and Exa are hosted only. For Tavily compatible APIs you run yourself, see self hosted Tavily alternatives.