TL;DR: For an agent that needs web search, Tavily gives 1,000 API credits a month with no credit card, and Exa gives $10 in credits a month plus a one-time $10 bonus. To turn known URLs into clean text, Firecrawl gives 1,000 credits a month but only 2 concurrent requests. WebScraping.AI gives 2,000 credits a month, and Bright Data gives 5,000 Web Unlocker requests a month for sites that block bots.
Two jobs that both get called “web access”
An agent that browses the web needs one of two things.
The first is search. The agent has a question and no URL. It sends a query and gets back ranked results, often with the page text already attached. Tavily and Exa do this.
The second is fetching. The agent has a URL, or a list of them, and needs the page as HTML, Markdown or JSON. The page may need JavaScript, a rotating proxy or a CAPTCHA solver before it loads. Firecrawl, WebScraping.AI, Bright Data, ScraperAPI, Zenscrape, Microlink and Apify do this.
Some tools do both. Firecrawl has a search endpoint next to its scrape and crawl endpoints. Even so, each tool here is best at one of the two jobs.
The tables below cover the free plan of each tool. The last column is the first paid step, because the free limit that stops an agent is usually the one the paid plan raises. All nine tools sit in the Web Scraping APIs category. Exa is also in Search. Firecrawl and Apify are also in Browser Automation.
Table 1: search APIs for agents
| Tool | Free allowance | Resets | Limit that matters | First paid step |
|---|---|---|---|---|
| Tavily | 1,000 API credits, no credit card | 1st day of each month | Credits per request vary: search, extract and crawl cost different amounts. Requests stop at zero. | Pay As You Go at $0.008 per credit, or Project at $30/month for 4,000 credits and higher rate limits |
| Exa | $10 in credits, up to 2,500 Lite searches, plus a one-time $10 onboarding bonus. No payment method. | 1st day of each month | 10 search queries a second and 50 agent concurrency | Pay as you go, with Search from $4 per 1,000 requests and up to 25 queries a second |
Tavily counts credits, Exa counts dollars
Tavily’s unit is the API credit. Search, extract and crawl requests use different amounts, and advanced requests use more. The balance resets on the first of the month, whatever day you signed up. Tavily’s pricing page also says the service is free for students.
Exa’s unit is the dollar. The $10 buys up to 2,500 of its Lite searches, but the same balance goes faster on Deep Search at $12 to $15 per 1,000 requests, or on Contents at $1 per 1,000 pages. The one-time $10 bonus arrives after you finish onboarding in the dashboard. After that the free balance is $10 a month.
Exa publishes a throughput limit for its free plan: 10 queries a second. Parallel agents can plan around that number. For a side-by-side of the two, see Exa vs Tavily. The Tavily alternatives page lists more search options.
Table 2: scraping and crawling APIs
| Tool | Free allowance | Resets | Limit that matters | First paid step |
|---|---|---|---|---|
| Firecrawl | 1,000 credits: 500 searches or 1,000 pages. No card. | Monthly | 2 concurrent requests, low rate limits | Hobby at $19/month: 5,000 credits, 5 concurrent requests |
| WebScraping.AI | 2,000 API credits. No card. | Monthly | 2 concurrent connections | Personal at $29/month: 250,000 credits, 10 concurrent requests |
| Bright Data (Web Unlocker) | 5,000 requests on the free tier | Monthly | n/a | Pay as you go at $1.50 per 1,000 requests, unlimited concurrency |
| ScraperAPI | Free trial with 5,000 API credits | One-time trial | n/a | n/a |
| Zenscrape | 1,000 credits on the Free plan | n/a | n/a | Small at $59.99/month: 250,000 credits, 10 concurrent requests |
| Microlink | 25 requests a day. No card. | Daily | The quota is daily only, with no monthly pool | Pro at $49/month: 46,000 requests a month |
| Apify | $5 of platform usage on the Free plan. No card. | Monthly billing cycle | Pay per compute unit at $0.20 | Starter at $19/month, with $19 of usage included, plus pay as you go |
Credits are not requests
Firecrawl’s own table shows the gap. 1,000 credits is 1,000 scraped pages or 500 searches. A search costs two credits. Check the cost of each call before you size a plan.
WebScraping.AI uses credits too, at 2,000 a month. Its 2 concurrent connections is the same cap as Firecrawl’s. One agent working through a queue can live with that. An agent that opens ten pages at once cannot.
Bright Data counts requests, not credits: 5,000 Web Unlocker requests a month. Its pricing page also marks the SERP API, Scraper APIs and Scraper Studio as having a free tier. Try it when a target site blocks ordinary scrapers. The paid rate is $1.50 per 1,000 requests, and the page says paid concurrency is unlimited.
The daily and the dollar plans
Microlink is the odd one out. Its 25 requests a day reset every day, so the free plan has no monthly pool to save up. A daily cap suits a small tool that makes a few calls on demand, such as a link-preview card or a screenshot. It does not suit a nightly crawl. Pro moves to a monthly quota of 46,000 requests.
Apify meters in dollars. The Free plan holds $5, and the page says unused credit expires at the end of each billing cycle. Past the free plan you do not pick a request count. You pay for the compute your Actor uses, at $0.20 per compute unit on Free and Starter. Run one Actor and read its cost before you plan around the $5.
ScraperAPI offers a free trial with 5,000 API credits. Zenscrape lists 1,000 credits on its Free plan. Check the vendor’s pricing page for the rest before you depend on either.
For a head-to-head between two of the fetch tools, see Firecrawl vs Tavily and Browse AI vs Firecrawl.
What the free tiers cannot do
Every plan here has a ceiling that an agent in production passes quickly. 1,000 credits is about 33 a day. Exa’s $10 gives up to 2,500 Lite searches a month, about 83 a day.
The concurrency caps matter more than the credit totals for interactive agents. With 2 concurrent requests, an agent that fans out hits the cap on its third parallel call. The cheapest in this article is Firecrawl’s Hobby at $19 a month, which raises the cap to 5. Apify’s Starter is also $19 a month.
A worked example: 50 searches and 200 pages a day
Credit totals are hard to compare in the abstract, so take one agent and run the numbers. It makes 50 web searches and fetches 200 pages a day. Over a 30-day month that is 1,500 searches and 6,000 page fetches. Each figure below uses the per-call cost the vendor publishes.
Tavily: A basic search costs 1 API credit, and an advanced search costs 2. Basic Extract costs 1 credit for every 5 successful URLs, and a failed extraction costs nothing. The agent uses 50 credits a day on search and 40 on extraction, so 90 a day. The 1,000 free credits last about 11 days.
Exa: Lite search costs $4 per 1,000 requests, so 1,500 searches cost $6. Contents costs $1 per 1,000 pages for each content type, so 6,000 pages of text cost $6. The total is $12 against a $10 monthly balance. The searches alone fit. In the first month, the $10 onboarding bonus covers the gap.
Firecrawl: Scrape costs 1 credit per page, and Search costs 2 credits per 10 results. The agent uses 100 credits a day on search and 200 on pages, so 300 a day. The 1,000 free credits last about 3 days. Hobby’s 5,000 credits last about 16 days.
WebScraping.AI: A request without JS rendering costs 1 credit on datacenter proxies. With JS rendering, the default, it costs 5. At 200 rendered pages a day the agent uses 1,000 credits a day, so the 2,000 free credits last 2 days. Turn rendering off for static pages, and the same 2,000 credits last 10 days.
Bright Data: The free tier counts requests. 6,000 fetches pass the 5,000-request allowance after about 25 days.
Microlink: The 25 daily requests cover 25 of the 200 pages.
The settings you choose per call matter as much as the vendor. Search depth on Tavily and JS rendering on WebScraping.AI can change the cost by two to five times. Read the cost table before you set the agent’s defaults.
What this list leaves out
Abstract’s web scraping API is not covered. No-code scrapers such as Browse AI are also left out. They target people who click rather than agents that call an API, and they belong in their own comparison.
Which one to start with
For an agent that answers questions with live web data, start with Tavily if you want a flat credit count and no card, or Exa if you want a published rate limit and a bonus in the first month.
For an agent that reads known pages, start with Firecrawl, which needs no card. Switch to WebScraping.AI when you need more credits a month. Choose Bright Data when pages block ordinary requests. Pick Microlink for previews and screenshots, and Apify when a ready-made Actor already scrapes the site you need.
Students should check Tavily first, because it is free for them. If you want a plan with no credit to run out, look at the always-free listings.