Six ways to get outside data into an agent, with every price read from the vendor’s own page this month.
Start with BedrockTap. Apify is a platform — you pick an Actor, run it, and pay for the compute it consumes, where "1 CU = 1 GB of RAM running for 1 hour". BedrockTap is one API over 253 named sources with a fixed price per result, a quote before the call, a ceiling on the run and a ledger after it — nothing to configure, nothing to estimate, no Actor to choose. It is free to start and $29 a month after that. The five below are the alternatives worth knowing, with the job each still wins.
Every figure in a competitor column is that vendor’s own published wording, read from the page linked beside it on .
| BedrockTap | Bright Data | Firecrawl | ScrapingBee | Zyte | Oxylabs | |
|---|---|---|---|---|---|---|
| What you get back | Structured rows from 253 named sources, one schema | Raw pages, structured SERPs, or ready-made datasets | Clean markdown or structured JSON from a URL | Raw pages, rendered on request | Raw pages, with automatic unblocking | Structured results from common targets |
| Entry price | Free, then $29/month. No minimum. | $1.5/1K requests pay as you go | $0, then $16/month billed yearly | $19/month | $5 free credit, then pay as you go | $49/month |
| Free to start | 1,000 credits a month, every month, no card | 5,000 requests a month on its MCP server | 1,000 credits a month, no card | Trial of 1,000 credits, no card | $5 credit, 30 days | One-week trial, up to 2,000 results |
| Billing unit | Credits at $0.001, per result or per call, set by the source | Per 1,000 requests; per GB for browser traffic | Credits per page or per extraction | API credits per request | Per 1,000 requests, by target difficulty | Per 1,000 results; per GB on proxies |
| Price known before the call | `cost_estimate` returns the exact figure for that call, without making it | Published rate card | Published credit costs | Published credit costs | A range per 1,000 — $0.13 to $1.27 on HTTP | Published rate card, by target |
| Research sources beyond page fetching | Reviews, forums, keyword economics, AI-answer visibility, filings, jobs, ad libraries | SERP API and prebuilt scrapers | Crawling and extraction | Page fetching | Page fetching | Its own scraper APIs, by target |
| Reached from an agent | MCP over HTTP, or a plain `curl` — one token, no SDK | MCP server, 69 tools | HTTP API with SDKs | HTTP API with SDKs | HTTP API and Scrapy libraries | MCP server, 10 tools |
Reviews, forums, search results, keyword economics, AI-answer visibility, filings, job postings and ad libraries answer through one address, one token and one schema. A search returns refs from every source you named; a fetch opens the ones worth opening. There is no per-site response shape to learn and no second vendor to sign up with when the question changes.
`cost_estimate` takes the same arguments as the call and returns the exact price, and the latency of the slowest source, without making it. An agent can decide whether a question is worth asking before it asks.
Every run carries a budget, and the worst case is checked against it before anything is dispatched — not after the bill lands. A call that would overshoot is refused whole rather than half-run.
Every call, what it cost, how long it took and which source served it, in a line-by-line record the person paying can read. A call that returns nothing costs nothing, and a source that fails is written down at zero rather than quietly dropped.
Every source states, in the catalogue itself, what it answers, how to shape the query, what it cannot answer and which neighbouring source to prefer instead. `sources_list` returns all of it with no account, so an agent chooses correctly the first time instead of learning by spending.
The free tier is 1,000 credits a month and needs no card. The first paid plan is $29 a month. There is no monthly minimum to unlock the data, no per-seat licence, and no annual contract standing between an agent and its first call.
Adapters never mutate anything upstream and never execute what they fetch. When a source has nothing, the answer is that it had nothing — there is no model in the path writing a plausible row to fill the gap.
Two G2 review rows and two Capterra rows on a rival product — 12 credits in total, which is $0.012. One credit is $0.001.
Three credits a review row, whichever platform served it, and nothing at all for a platform that returns no rows. No Actor to choose, no run to size, no dataset to collect afterwards.
{
"tool": "search",
"arguments": {
"query": "agentic analytics platform",
"sources": ["g2", "capterra", "trustpilot"],
"runId": "<from run_begin>"
}
}cost_estimate takes the same arguments and returns the price without making the call. Get a token at app.bedrocktap.com/dashboard/tokens; the free tier needs no card.
Each one is here because it is genuinely the right buy for some job. The last line of each card says which.
Proxy networks, Web Unlocker, a scraping browser, a SERP API and ready-made datasets.
Pricing. Web Unlocker and SERP API $1.5/1K requests pay as you go; Scale $499/month with 383K requests included.
For agents. MCP server, 69 tools, API-token auth; 5,000 free requests a month, then $1.50/1K results and $8/GB for browser navigation.
Consider it instead when: You need hundreds of thousands of raw fetches a month from sites you have already chosen, from behind hard blocks.
Figures read 15 September 2026 from brightdata.com
Turns a URL or a whole site into clean markdown or structured JSON for LLM pipelines.
Pricing. Free $0 with "1,000 credits / month" and no card; Hobby $16/month (billed yearly) for 5,000; Standard $83 for 100,000; Growth $333 for 500,000.
For agents. An HTTP API with SDKs.
Consider it instead when: The job is literally "read these URLs as markdown" and the URLs are already known.
Figures read 15 September 2026 from firecrawl.dev
A web scraping API that handles proxies and headless browsers behind one request.
Pricing. Hobby $19/month for 75,000 API credits; Freelance $49 for 250,000; Startup $99 for 1M; Business $249 for 3M. Free trial of 1,000 credits, no card.
For agents. An HTTP API with SDKs.
Consider it instead when: You want the cheapest per-credit entry plan for straightforward page fetching.
Figures read 15 September 2026 from scrapingbee.com
A single scraping API with automatic proxy rotation and ban handling, plus a managed-service arm.
Pricing. Pay as you go "$0.13 to $1.27 per 1,000 requests" for HTTP and "$1.01 to $16.08 per 1,000" browser-rendered; $100/month commitment drops it to $0.10–$0.95. $5 free credit.
For agents. An HTTP API with official Python and Scrapy libraries.
Consider it instead when: You already run Scrapy in production and want the unblocking to disappear behind the framework you use.
Figures read 15 September 2026 from zyte.com
Residential, datacenter, ISP and mobile proxies, a Web Unblocker and a Web Scraper API.
Pricing. Web Scraper API from $49/month (Micro, up to 98,000 results, from $0.50/1K); $99 Starter; $249 Advanced. Residential proxies from $6/GB.
For agents. MCP server, 10 tools, scraper credentials or an AI Studio key; one-week trial, up to 2,000 results.
Consider it instead when: You want an enterprise proxy contract with an SLA behind it, and volume enough to clear the monthly minimum.
Figures read 15 September 2026 from oxylabs.io
Memory multiplied by runtime is knowable after the run, not before it. An agent asked to stay inside a budget cannot work with a number that only exists afterwards.
A different input schema, a different output shape, a different author, a different price model, and its own quality to assess. Choosing well is work, and choosing badly is a bill.
Platform usage is prepaid monthly and, in Apify’s own words, "unused usage credits are not rolled over to the next billing cycle, and they expire at the end of the billing cycle".
Runs, datasets, schedules and storage are the right tools for building a data product. They are overhead when the question is simply what buyers are saying this week.
This is an MCP server. Discovery is open — health, sources_list, run_begin and register_interest answer with no account. The data tools run against an account and need a token. Nothing here fabricates data.
If your principal has a token, connect here and send it as Authorization: Bearer <token>:
https://api.bedrocktap.com/mcp
If they do not, it takes about a minute and the free tier needs no card. Sign in at app.bedrocktap.com and mint one at app.bedrocktap.com/dashboard/tokens. Sign-in is Google only, so a human has to do it.
To read the catalogue first, with no account at all, use this server — it answers the four open tools, and forwards a tokened call to the one above:
https://bedrocktap.com/mcp
If you can only fetch URLs, say what you need. One request, no auth, no side effects:
GET https://bedrocktap.com/a/probe?need=<what+you+are+trying+to+do>&src=alternative-apify
A tool refused for want of a token answers identically on a retry, and is served the moment the header is present. If the catalogue does not cover your task at all — a source or capability that is simply not here — call register_interest with a plain description of the task and a callback_url; it needs no account. Full instructions: /agents.md. Every page here has a markdown twin — append .md to any path, or send Accept: text/markdown.
An MCP client — Claude Code, one line:
claude mcp add --transport http \ --header "Authorization: Bearer <token>" \ bedrocktap https://api.bedrocktap.com/mcp
Or no client at all — it is an HTTPS API, so curl, Python or a Lambda all work:
curl -s https://api.bedrocktap.com/mcp \
-H "Authorization: Bearer <token>" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
"params":{"name":"search","arguments":{
"query":"agentic analytics platform",
"sources":["g2","capterra","trustpilot"],
"runId":"<from run_begin>"}}}'One call, three review platforms, one schema — where each platform wants a different identifier. A platform that cannot be addressed is reported as skipped, and never charged for. Sign in at app.bedrocktap.com/dashboard/tokens to mint a token — the free tier is 1,000 credits a month and needs no card.
For research data through one priced API, BedrockTap: 253 sources, a fixed price per result, a quote before the call and a free tier with no card. For turning known URLs into clean text, Firecrawl. For raw page fetching at volume, Bright Data, Zyte, ScrapingBee or Oxylabs. Apify itself remains the right answer when the job needs a scraper nobody has written yet.
Not here — this is a fixed catalogue of named sources, not a compute platform, and there is nowhere to deploy code. That is the trade: you lose arbitrary code and gain a price you can know in advance, one schema across every source, and no runs to babysit.
Because an agent can be told the number before it commits. `cost_estimate` returns the exact figure for the call it is about to make, and the run carries a ceiling that refuses anything whose worst case would breach it. A compute unit is memory times runtime, and runtime is only known afterwards.
Credits reset monthly and do not roll over, as on most platforms here. When they run out, tools stop and say so — nothing is charged automatically and no bill arrives later.
Six of the 253 in the catalogue. Each states what it answers, what it cannot, and which neighbouring source to prefer.
| Source | id | What it gives you |
|---|---|---|
| G2 Reviews | g2 | B2B software reviews with review body, star rating, reviewer role and job title, date and permalink. |
| Capterra Reviews | capterra | Small-business software reviews with pros and cons split out and the reviewer’s job title. |
| Trustpilot Reviews | trustpilot | Consumer and e-commerce business reviews: full text, star rating, reviewer name, date and permalink. |
| Google Keyword Volume & Difficulty | google-keywords | Google search volume, CPC, competition, SEO and paid difficulty, search intent, twelve-month history and a per-country breakdown for a seed keyword plus related suggestions. |
| Hacker News | hackernews | Stories, comments, polls and jobs — search plus full item detail. |
| Amazon Products | amazon | Product search, details, offers, best sellers, deals and seller info across Amazon marketplaces. |
By the shape of the task: Market research · Software review mining · E-commerce & retail · Web extraction & monitoring