Apify MCP Server: Actors as AI Tools, and What They Cost
Yes — Apify has an official MCP server, and it’s one of the more ambitious ones out there. Instead of exposing a fixed handful of tools for one product, Apify’s server is a doorway onto the whole Apify Store: your AI client can search for an Actor, read its input schema, add it as a tool mid-conversation, and run it. Apify’s own MCP page puts it as “One MCP server, access to thousands of tools,” and says a “Single Apify API key unlocks 7,000+ Actors and MCP servers.”
That design is the interesting part and the part worth being careful about. A server whose tool list grows on demand is genuinely powerful, and Actors cost real money to run — so an AI that can add and invoke scrapers on its own is spending your credit balance while it works.
Apify is one of many — the complete list of MCP servers covers every app that ships one, grouped by category.
What the Apify MCP server actually exposes
Model Context Protocol is the open standard that lets an AI client — Claude, ChatGPT, Cursor, VS Code — talk to an outside system through a shared interface. Apify maintains its server itself, in the open, at apify/apify-mcp-server.
By default you don’t get 7,000 tools dumped into your context. Apify’s docs state that “by default, the MCP server loads essential tools for Actor discovery, documentation search, and the RAG Web Browser Actor.” In practice that default set is:
search-actors— find an Actor in the Apify Store by what you’re trying to do.fetch-actor-details— pull an Actor’s input schema and docs so the model knows how to call it correctly.search-apify-docs/fetch-apify-docs— search and read Apify’s own documentation, which is why this server is useful to coding assistants even before you scrape anything.apify/rag-web-browser— search the web and return page content, loaded by default.
Beyond that, call-actor runs an Actor and returns results, add-actor promotes a discovered Actor into a live tool for the rest of the session, and helpers like get-actor-run, get-dataset-items, get-key-value-store-record, and abort-actor-run handle runs and storage.
Tool count is dynamic, not fixed. That’s the headline behavior: the docs describe “the ability to discover and use new Actors on demand,” and note that “each discovered Actor becomes immediately available for future use in the conversation.” Support for adding tools mid-session depends on your client handling notifications/tools/list_changed — it works in clients like Claude.ai and VS Code, and degrades to “restart with the Actor preloaded” in ones that don’t.
Connecting: hosted or local
Apify’s docs are direct about the two paths: “You can connect to the Apify MCP server in two ways: use our hosted service for a quick and easy setup using Streamable HTTP with OAuth, or run the server locally for development and testing using local stdio.”
Hosted (recommended for most people). Point your client at https://mcp.apify.com. Auth is either browser-based OAuth on first connect, or an Authorization: Bearer <APIFY_TOKEN> header with a token from API & Integrations in Apify Console. You can pin the tool set with a query parameter rather than accepting defaults, e.g. https://mcp.apify.com?tools=actors,docs,apify/rag-web-browser. There’s a configurator UI at mcp.apify.com that builds the URL and client config for you.
Local. npx @apify/actors-mcp-server over stdio, with APIFY_TOKEN in the environment. Same tools, useful when you’re developing against it.
Two details that save time later. Unauthenticated access exists but is deliberately thin — anonymous connections get only search-actors, fetch-actor-details, search-apify-docs, and fetch-apify-docs, i.e. discovery and reading, no running. And there’s a documented ceiling: “The Apify MCP server allows up to 30 requests per second per user.”
Separately, Apify lets any Actor be its own MCP server. An Actor running in Standby mode exposes a Streamable HTTP endpoint at /mcp on its own .apify.actor host, which is how the MCP servers listed in the Apify Store work. So “Apify MCP” can mean the platform-wide server at mcp.apify.com or one purpose-built Actor — worth knowing which one a tutorial is talking about.
If you’re wiring this into a specific client rather than reading generally, MCP servers in ChatGPT and MCP servers in Codex cover the client-side setup and its quirks.
The cost question nobody puts in the README
This matters more for Apify than for a typical MCP server, because Apify Actors are usage-priced. Running one is not a free API read.
Apify Actors bill on a few models: pay per event, where “you pay for specific events the Actor creator defines, such as generating a single result or starting the Actor”; pay per usage, charged against platform resources — compute units, storage operations, proxy traffic; and a legacy rental model being phased out. The free plan comes with $5 of platform credit per billing cycle at $0.20 per compute unit, and new runs are blocked once it’s spent.
Now put an AI in front of that. A model that can search-actors, add-actor, and call-actor in a single turn can start a paid job because it inferred you wanted one. A per-result Actor priced at, say, a few dollars per thousand records will burn a free month’s credit in one enthusiastic scrape. Practical guardrails: keep the tool set explicit via the tools= parameter instead of letting discovery run wide, set usage limits in Apify Console rather than trusting the model to be frugal, and check an Actor’s Pricing tab before you let a client call it repeatedly.
Where the Apify MCP stops
The limits here are the protocol’s, not Apify’s — Apify has implemented about as much as MCP allows.
There is no such thing as an MCP trigger. The protocol's own working group states plainly that clients find out about server-side changes by polling for them, and a real event mechanism remains unshipped.
That’s the shape of the thing. The Apify MCP server answers when asked. It does not watch anything, and it does not run on a clock. Which means:
- No event triggers. A dataset finishing, a scheduled Actor completing, a new item landing in a key-value store — none of these can start anything through MCP. You find out by asking.
- It hands you results, not a pipeline. Getting scraped data into a sheet, a CRM, and a Slack message means the model doing three more things correctly in the same conversation, with a separate MCP server authed for each.
- The blast radius is your account. The token you paste grants Actor execution under your own plan and your own balance.
Both ChatGPT and Claude will now do work on a schedule, unattended. Neither reacts to events. A task that runs hourly still finds out an hour late, and one that runs each morning finds out tomorrow — the trigger is the clock, never the thing you actually care about.
Making scraped data go somewhere on its own
Scraping is rarely the goal. The goal is usually “when new results show up, do something with them” — and that’s the part MCP hands back to you.
Carly connects to Apify natively, alongside ~260 other apps, plus anything with a public API through your own key. The difference is that Carly’s workflows are triggered and scheduled, so an Apify run and everything downstream of it happens without a conversation being open: run a scrape on a schedule, take the dataset, dedupe it against what you already have, enrich the rows, write them to a sheet or your CRM, and post a summary to Slack — as one automatic sequence rather than a chat you have to remember to start. Non-AI steps (moving, matching, routing) are free and unlimited; AI steps like summarizing or deciding start at $35/month.
FAQ
Does Apify have an official MCP server?
Yes. Apify builds and maintains it, open source at apify/apify-mcp-server, hosted at https://mcp.apify.com over Streamable HTTP, with a local stdio option via npx @apify/actors-mcp-server.
How does Apify MCP authentication work?
Two ways on the hosted server: OAuth, where you sign in through the browser on first connection and never paste a token, or an Authorization: Bearer <APIFY_TOKEN> header using a token from Apify Console’s API & Integrations section. Local runs read APIFY_TOKEN from the environment. Anonymous connections work but are limited to Actor discovery and docs search — no running Actors.
How many tools does the Apify MCP server have? It varies by session. Defaults cover Actor discovery, Apify docs search, and the RAG Web Browser Actor; from there the model can search the Store and add Actors as tools on demand, against a catalog Apify describes as 7,000+ Actors and MCP servers.
Does using Apify through MCP cost money? Running Actors does, yes. Actors bill per event or per platform usage (compute units, storage, proxy), independent of MCP — MCP is just the calling mechanism. Free accounts get $5 of credit per cycle. Because an AI can discover and invoke Actors on its own, set usage limits in Apify Console before connecting a client.
Can the Apify MCP server run a scrape on a schedule? Not through MCP. MCP has no scheduler and no triggers — a call happens because you asked for it in that moment. Apify’s own platform schedules Actor runs, but reacting to the results across other apps needs a workflow tool. That’s what Carly does: an Apify run on a trigger or timer, with the output routed onward automatically.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
