mrkeyoor.com_
Tue 01 Sept 16:14 UTC
WebMCP servermeasured 19 Aug 2026

Apify Actors

Apify Actors is an MCP server from Apify that exposes its cloud scraping and automation actors to an AI client over stdio. You point a client at @apify/actors-mcp-server, it authenticates to your Apify account, and the actors you pick become callable tools. It runs your scrapes on Apify's platform, not locally, and bills your account.

BootDID NOT BOOTWe started it and it never completed initialize: exited with code 1 before initialize completed
Tools0tools/list payload 0 bytes
Context tax~0estimated tokens for the whole tools/list, chars/4 estimate
Resources / prompts / counts returned by resources/list and prompts/list
Measured19 Aug 2026one run of the harness, clean environment, no credentials · how we test

How we started it

npx -y @apify/actors-mcp-server
Verdict

We could not get a tool list out of it, because it refuses to start without credentials: our clean-env run exited with code 1 before initialize completed (measured 2026-08-19), with stderr saying APIFY_TOKEN is required but not set in the environment variables or in ~/.apify/auth.json. Use it if you already pay Apify and want your agent to reach the actors you run there anyway; the token requirement is honest, since there is no useful local mode to fall back on. Skip it if you were hoping for a scraper you can try before you have an account, and note that everything this server does costs Apify credits, so a chatty agent is a billing surface as much as a context one.

Use it if

You already have an Apify account with actors you run, and you want an agent to trigger them instead of you clicking Run.
Your scraping targets need residential proxies and retries that you do not want to host yourself.
You want the crawl to happen on someone else's machine, with the results fetched back, rather than inside your laptop's Node process.
You are comfortable with a token in your MCP client config that can spend money.

Skip it if

You want to evaluate the tool surface before signing up; there is nothing to look at until APIFY_TOKEN is set.
You are running MCP servers in a sandbox or CI job where injecting a paid-account token is not acceptable.
Your scraping is a handful of static pages a fetch tool already handles, and per-run credits are a worse deal than a plain HTTP request.
You need deterministic, auditable costs per agent turn, since actor runs bill by compute and an agent decides how many to launch.
You want an offline or self-hosted crawler; this is a remote-execution front end for a hosted platform.

The bill for your context window

There is no Context Tax to report here: the server never answered tools/list, so toolsCount and estTokens are zero for the reason that it exited, not because it is lean. What we can say is what to expect once you supply a token. The tool surface is dynamic, driven by which actors you enable, so this is one of the few servers where the context cost is a dial you control rather than a fixed number you inherit. Enable a single actor and the payload stays small; enable a shelf of them and you can outweigh the heaviest fixed-tool server in this directory without noticing you did it. Whatever you enable, the estimate method is the same chars/4 count we apply to every server here, so you can compare once you have your own list.

Setup reality

One env var stands between you and a tool list, and the harness found exactly that one: APIFY_TOKEN, scraped from the server's own error output. It also accepts a token sitting in ~/.apify/auth.json, which is what the Apify CLI writes when you log in, so on a developer machine that has used the CLI it will often appear to need no configuration at all. That is worth knowing before you copy a working config to a server and watch it die. Transport is stdio via npx, no build step, and the install pulled a deprecation warning about lodash.isequal on the way past. Which actors show up as tools depends on your account and on the actor selection you pass, so two people running the same command get different tool surfaces.

APIFY_TOKEN

Questions people ask

Why did the Apify MCP server fail to start in your test?

We run every server as an unprivileged user with a clean environment and no secrets, which is deliberate. Apify's server checks for APIFY_TOKEN in the environment or in ~/.apify/auth.json, finds neither, prints an error and exits with code 1 before the MCP initialize handshake.

Can I use the Apify MCP server without an Apify account?

No. The token is not optional and there is no local mode, because the actors run on Apify's infrastructure rather than on your machine. Signing up is the first setup step, not an upsell you can defer.

How many tools does it expose?

That depends on you. The tool list is built from the actors you enable for the server, so the count and the context cost vary per account. We could not measure a number for it, since our run never reached tools/list.

Does running it cost money?

Yes, in Apify platform credits, on the same meter as running those actors by hand. The difference is that an agent decides how often to launch them, so set actor limits before you hand the token over.

Other web servers we started

ServerVendorBootsContext tax
FetchAnthropic (reference)~276 tokens
FirecrawlFirecrawl~9,077 tokens

How this page is made: the server is spawned as an unprivileged user with a clean environment and no credentials, then asked for its tools, resources and prompts over stdio. Token figures are estimates at four characters per token, not a tokenizer count. One run, one machine. Corrections: contact the desk.

← All measured MCP servers