Best Apify Actors to expose as MCP tools
Pin four Actors and stop. apify/rag-web-browser searches and reads in one call and carries no Actor fee. apify/web-fetch pulls one known URL, from $1.50 per 1,000 fetches. apify/google-search-scraper returns structured SERPs at $0.0045 per result page. apify/website-content-crawler belongs in a batch job, not a chat turn.
Every price and user count below was read from Apify's Store API and Store pages on 2026-09-09. This page narrows down to the four Actors worth exposing as MCP tools; for the wider picture across every category, see best Apify Actors. The setup, the auth flow and the 30 req/s rate cap live in Apify MCP vs the API.
Why do most Apify Actors make bad MCP tools?
Apify publishes the Store as 70,000+ ready-to-run tools (apify.com/store, read 2026-09-09). Nearly all of them can be called over MCP. Nearly none of them should be.
A tool call has to fit inside a conversation and survive being re-read. Point Website Content Crawler at a mid-size documentation site and it returns tens of thousands of tokens of markdown. That payload then sits in the context window for every turn that follows. The run also has to finish while the client waits. And the model, not you, fills in the inputs: an Actor with twenty configuration fields gives it twenty chances to guess wrong. You pay for the run either way.
That is the whole selection problem. Four questions settle it:
- How big is one response? A few KB of markdown or a handful of JSON fields. Not a 50 KB HTML-to-markdown dump.
- How long is one run? Under about 30 seconds, or the assistant stalls mid-answer. Prefer HTTP-based scrapers over headless browsers where both work.
- How narrow is the input? One
querystring or oneurlis ideal. - What does a retry cost? Agents retry. An Actor with an expensive per-event price is a bad thing to hand a model that guesses.
Which four Actors are worth pinning, and what does each call cost?
| Actor | Returns | Actor fee | Pin it for |
|---|---|---|---|
| apify/rag-web-browser | Markdown per result, plus title, URL, status, language | None. Platform usage only (pricingModel: FREE, Store API, 2026-09-09) | Open-ended research inside the conversation |
| apify/web-fetch | One page as markdown, text, HTML or links, plus metadata | From $1.50 / 1,000 fetches, pay-per-event ($0.0015 per fetch) | "Read this URL for me" when the URL is already known |
| apify/google-search-scraper | Structured SERP results | $0.0045 per result page scraped, Free-tier rate | Rankings, competitor checks, anything that needs the SERP itself |
| apify/website-content-crawler | Whole sites as clean markdown | None. Platform usage: about $0.2 / 1,000 pages raw HTTP, $0.5–$5 / 1,000 headless | Batch RAG ingestion, run outside the agent loop |
Actor fees read from api.apify.com/v2/store and each Actor's Store page on 2026-09-09. Platform compute is billed on top at $0.20 per compute unit on the Free and Starter plans (apify.com/pricing, read 2026-09-09).
RAG Web Browser is the one to pin first
It takes a query or a URL, runs the search, fetches the pages, strips the chrome, and hands back small markdown chunks with metadata. Apify enables it by default on the hosted server, which is a fair signal about what it was built for. It is also the most-used of the four in this role: 169,448 total users, 30,683 monthly, 4.71 stars, 99.7% success rate (Store page, 2026-09-09).
Keep maxResults at or below 10. The Actor's own README says to "limit the results to 10 using the maxResults parameter", because past ten it has to page through several Google results pages and the run slows down (Store page, read 2026-09-09). The field itself defaults to 3 and accepts up to 100, so the cap is yours to set — and ten results is already several thousand tokens.
Web Fetch is the cheaper answer when you already have the URL
Web Fetch does one job: download a page, PDF, or image and convert it. It handles JS rendering and bot challenges, and failed requests are not charged. At $0.0015 per fetch plus a $0.00005 start event, a hundred agent lookups cost about 15 cents in Actor fees.
If the assistant needs to find the page rather than read a page you named, RAG Web Browser is the right tool and Web Fetch is the wrong one.
Google Search Results Scraper carries a trap worth knowing
The base rate is cheap: $0.0045 per scraped result page. Its charge events also include add-ons. Google AI Mode, Perplexity, ChatGPT, Copilot and Gemini results bill at $0.20 each; business-lead enrichment and email verification at $0.10 each (Store API, read 2026-09-09).
That is up to 44× the base rate on one event, decided by an input the model filled in. Pin this Actor when you need SERP data. Set an account spend limit before you do.
Website Content Crawler is the one to keep out of the tool list
It is the strongest crawler of the four: sitemap-aware, JS rendering, deduplication, 155,194 users at 4.61 stars and a 99.2% success rate (Store page, 2026-09-09). It is also the one most likely to wreck an agent conversation. A crawl of a mid-size docs site returns tens of thousands of tokens in a single tool response.
Run it as a scheduled job, push the output to a vector store, and let the agent query the store. That is the standard pattern, and our data for AI and RAG guide walks through it. If you must expose it live, hard-cap maxCrawlPages at 5–20.
What else should stay out of the tool list?
Anything that paginates into thousands of items. Social and marketplace scrapers are built to fill a dataset, not to answer a question in a chat turn. Point the agent at the dataset afterwards instead. Our best AI and RAG Actors roundup covers those.
Actors you pinned by slug and never re-checked. A pinned tool that stops existing does not error; it simply is not there. The clearest example is apify/actors-mcp-server. It used to be an Actor you ran. It is not one now: apify.com/apify/actors-mcp-server redirects to mcp.apify.com and the API returns 404 for that Actor ID (checked 2026-09-09). The hosted endpoint replaced it, and the npm package @apify/actors-mcp-server still exists for local stdio use. Any config or tutorial that names the Actor is describing something that no longer exists.
Is Firecrawl's MCP server the better choice?
Sometimes, and it is worth saying plainly.
Firecrawl's MCP server exposes a purpose-built set of around fifteen named tools — firecrawl_scrape, firecrawl_search, firecrawl_map, firecrawl_crawl, firecrawl_parse, firecrawl_agent and the rest (docs.firecrawl.dev, read 2026-09-09). Billing is one unit: 1 credit = 1 page on a basic scrape, crawl or map, and search at 2 credits per 10 results (firecrawl.dev/pricing, read 2026-09-09). The free tier is 1,000 credits a month on that same page, and Hobby is $16/month billed annually for 5,000.
If your agent only ever needs to search, scrape and crawl generic web pages, Firecrawl is the simpler buy. One credit unit forecasts better than compute units plus per-event Actor fees. You also skip the question of which tools to load.
Apify earns its place when a generic crawler is not enough. A specific site's structured fields, a SERP with ads and AI Overview blocks, a marketplace listing, a scheduled pipeline on the same account. Its free tier is $5 of monthly usage rather than a credit count (apify.com/pricing, read 2026-09-09). That is harder to compare and easier to run down without noticing.
How do you stop an agent from spending your credit?
- Set an account spend limit in the Console before you connect anything. This is the only control that holds when a model retries in a loop.
- Keep the tool list to three or four entries. Every tool definition is serialized into the model's context on every turn, and a model that cannot see an expensive Actor cannot call it.
- Cap the arguments, not just the tools.
maxResults≤ 10 on RAG Web Browser,maxCrawlPageson Website Content Crawler. - Start on the free plan. It includes $5/month subject to change · verified 2026-09-09 of usage credit, which is enough to test a tool list and not enough to fund a runaway. Details in our Apify free plan breakdown.
One conflict to plan around. Apify's own pages disagree on how many runs a free account may execute at once. apify.com/pricing says 5 concurrent runs; docs.apify.com/platform/limits says 25. Both read 2026-09-09. We have not picked a side, because we cannot see which one billing enforces. Assume 5 when an agent fans tool calls out in parallel. If the real ceiling is 25 you lose nothing; if it is 5 you avoid a queue you did not plan for.
What changed in 2026 that breaks older configs
| Date | Change | Effect on a tool list |
|---|---|---|
| 1 Apr 2026 | SSE transport removed; Streamable HTTP only | A pre-April client config fails with a transport error, not a clear message |
| by Sep 2026 | apify/actors-mcp-server no longer exists as an Actor; apify.com/apify/actors-mcp-server redirects to mcp.apify.com (checked 2026-09-09) | Any tutorial telling you to add that Actor is describing a dead slug |
| 2026-09-09 | apify/rag-web-browser and apify/website-content-crawler both read as pricingModel: FREE in the Store API | No Actor fee on either; compute is still billed |
If you are copying a config from a tutorial written before April 2026, the transport line and the Actor slug in it are both wrong.
Create a free Apify account, set a spend limit, then point your client at https://mcp.apify.com?tools=actors,docs,apify/rag-web-browser.
Worried about cost? RAG Web Browser has no Actor fee at all, so a first afternoon of testing draws only against the $5 of monthly platform credit on the free plan.
apify/rag-web-browser. It searches and reads in one call, returns a few KB of markdown per result instead of a full page dump, and carries no Actor fee — the Store API reports pricingModel FREE as of 2026-09-09, so you pay Apify platform compute only. Apify also enables it by default on the hosted server.
It depends on the Actor. RAG Web Browser and Website Content Crawler charge no Actor fee, so you pay platform compute at $0.20 per compute unit. Web Fetch is $0.0015 per fetch, or $1.50 per 1,000. Google Search Results Scraper is $0.0045 per result page on the Free tier, with optional add-on events at $0.10–$0.20 each. All read 2026-09-09.
Because it is not an Actor any more. apify.com/apify/actors-mcp-server redirects to mcp.apify.com and the API returns 404 for that Actor ID (checked 2026-09-09). The hosted endpoint took its place. The npm package @apify/actors-mcp-server is unaffected and is still the local stdio server.
Three or four. Every tool definition is sent to the model on every turn, and the default configuration already gives it a search tool over a Store that Apify publishes as 70,000+ Actors. Naming specific Actors in the tools query string swaps that search box for explicit input schemas, which is what makes tool calls reliable.
For generic scrape, search and crawl work, Firecrawl is simpler: about fifteen purpose-built tools and one billing unit, 1 credit per page, 1,000 free credits a month (docs.firecrawl.dev and firecrawl.dev/pricing, read 2026-09-09). Apify wins when the agent needs a specific site's structured fields, SERP data, or a scheduled pipeline on the same account.
Yes, and the usual cause is an expensive per-event input the model chose. Google Search Results Scraper's add-on events cost $0.20 each against a $0.0045 base rate. Set an account spend limit first, keep the tool list short, and cap maxResults and maxCrawlPages in the tool arguments.
Common mistakes and fixes
No tools appear in Claude or Cursor after connecting.
The client connected but did not enumerate tools. Confirm the token at console.apify.com/account/integrations, then restart the client. If you pinned tools explicitly (`tools=apify/rag-web-browser,apify/web-fetch`), check every slug: a typo hides the tool silently instead of erroring.
The agent picks the wrong Actor or loops through discovery.
Default mode gives the model a search tool over the whole Store, which Apify publishes as 70,000+ Actors (apify.com/store, read 2026-09-09). Name three or four Actors in the `tools` query string so the model sees explicit input schemas instead of a search box.
One tool call returned a huge response and burned the context window.
Website Content Crawler and full-site crawlers return megabytes of markdown. Cap `maxCrawlPages` and `maxResults` in the tool arguments, or swap to apify/rag-web-browser, which returns a few KB of markdown per result.
A pinned Actor disappeared from the tool list after working for months.
Pinning is by slug, so a deprecated or renamed Actor becomes a missing tool with no error. Re-check the slug at apify.com/<user>/<actor>. The best-known case: apify/actors-mcp-server is no longer an Actor at all.