Is Apify worth the 20% cut for MCP servers?

For a small team or solo builder, usually yes: the 20% replaces building and operating metering, billing, payouts, fraud handling, and hosting. Self-hosting wins when you have engineering capacity, enterprise buyers who want invoices, or need custom auth.

What does self-hosting an MCP server require?

A server that speaks Streamable HTTP, your own authentication (API keys or OAuth), per-call metering, a billing integration such as Stripe, uptime monitoring, and your own discovery/marketing since no marketplace lists you.

Can you do both at once?

Yes. Nothing stops you from listing on Apify for discovery and pay-per-call billing while also offering a self-hosted or enterprise deployment of the same server for bigger customers.

apify vs self-hosting your mcp server: the real trade

elisabeth hitz · july 27, 2026 · 6 min read

apify hosts your MCP server, meters every call, bills the buyer, handles payouts, and lists you in a marketplace, for 20% of revenue plus real platform constraints. self-hosting keeps 100% and full control, but you build authentication, metering, billing, and discovery yourself. i run 19 servers on apify; here's the comparison i wish i'd read first.

this is the build-vs-buy spoke of how to sell mcp servers on apify. i'll declare my position up front: for a solo operator, i chose the platform, and i'd choose it again. the interesting part is where that choice flips.

the comparison, side by side

dimension	apify	self-hosted
billing + metering	built in, per event, buyer's account charged	you build it (stripe metered billing, usage records, dunning)
your revenue share	80% (platform keeps 20%)	100% minus stripe's roughly 3% + infrastructure
authentication	apify tokens, zero work	your own API keys or OAuth, plus rotation and abuse handling
hosting + scaling	standby containers, managed	your servers, your pager
discovery	store listing, SEO'd marketplace pages	none; every visitor is one you earned yourself
price agility	locked for one month per change	change whenever you like
exit ramp	paid actors can't simply unpublish; go free first, then wait	turn it off whenever you like
platform risk	daily automated tests can deprecate you; rules change	none, but all operational risk is yours
enterprise buyers	awkward: they want invoices and DPAs, not marketplace billing	natural fit for contracts and custom terms

when apify wins

you're small, the buyers are individual builders or agents paying cents per call, and your scarcest resource is time. the 20% buys the entire revenue back office. i priced that trade against my own situation honestly: building metered billing around stripe, key management, and monitoring for 19 servers would have consumed weeks i spent shipping instead. the mechanics of what the platform demands in return are documented in the deployment guide.

when self-hosting wins

enterprise motion. if the buyer is a company that needs an MSA, a DPA, and an invoice, marketplace billing is friction, not convenience.
price experimentation. the one-month lock (detailed in the pricing post) makes rapid testing impossible on the platform.
heavy or stateful workloads. the standby model fits stateless tool calls; long-running jobs with big memory footprints get expensive and awkward.
you already run infrastructure. if there's an ops team anyway, the marginal cost of one more service is small and the 20% starts looking like rent.

the third option nobody mentions: both

the models aren't exclusive. the store listing can be the discovery and self-serve tier, cents per call, zero sales conversation, while the same server ships as a private or self-hosted deployment for bigger customers on contract terms. the listing then does double duty as a live, testable demo for the enterprise conversation. that's the shape my catalog is heading toward: the store handles strangers, and the direct channel handles relationships, which is where my actual services business lives anyway.

my verdict

start on the platform. the day one buyer wants an invoice bigger than your monthly store revenue, add the self-hosted tier. reversing that order means spending your launch energy building billing infrastructure for customers who don't exist yet.

frequently asked questions

is the 20% cut worth it?

for a solo builder, usually. it replaces metering, billing, payouts, fraud, and hosting. it stops being worth it when you have engineering capacity to spare or enterprise buyers who need contracts.

what would self-hosting actually require?

streamable HTTP serving, your own auth, per-call usage metering wired into stripe, monitoring, and all of your own marketing, because no marketplace page exists for you.

does listing on apify lock me in?

partially. paid actors can't simply unpublish (free first, then wait), and prices lock monthly. your code stays yours though; the same server can run anywhere that speaks streamable HTTP.

whichever side you pick, the tooling discipline is identical: spec the tools, build with AI, verify everything before it touches a buyer. that workflow lives in my AI builder toolkit at closermethod.com/skills.

go deeper

sources: apify monetization and standby documentation; stripe pricing page for the roughly 3% processing comparison; the 20% margin, one-month pricing lock, unpublish restriction, and daily-test behavior verified first-hand on the live apify account, may through july 2026. written by elisabeth hitz, certified in anthropic's ai fluency program (framework & foundations, and ai capabilities & limitations), plus claude 101 and claude cowork.