Who should build it in-house?

A team with spare senior engineering capacity, a simple internal use case, and someone who will own the 2026-07-28 migration and the next one. If all three are true, build it. You will spend two or three engineer-weeks and own the result, which is cheaper than any vendor for a tool you will rarely change.

A team whose engineers are the bottleneck, whose server faces customers, or who does not want to own protocol migrations. The maintenance is the cost people forget: a server built in a weekend still needs someone awake when the spec revises.

What does "buy" actually cost over two years?

For a product server, roughly $4,500 up front and $700 a month, which is $21,300 across 24 months with migrations included. In-house is the loaded cost of the build plus every future migration you will do yourself, which is cheap until it is not.

Is there a wrong answer here?

Yes: buying a generic hosted server for a use case that needed a curated one, or building in-house and then not maintaining it. A server nobody maintains is worse than no server, whichever way you acquired it.

Compare

Build vs buy an MCP server

For roughly half of readers, the answer is build it yourself. If you have spare senior engineering time, a simple internal use case, and someone who will own the 2026-07-28 migration, in-house wins on cost and control. Buy when your engineers are the bottleneck, the server faces customers, or you do not want to own protocol migrations. If you are a two-person startup with idle senior devs, this is not Portkeel.

Side by side

Build in-house versus buy: what each optimises for
Dimension	Build in-house	Buy (Portkeel)
Up-front cost	2–3 engineer-weeks	$2,000–$4,500
Ongoing	Your team, unbudgeted	$250–$700/mo
Spec migrations	You own them	Included
Control	Total	Shared, documented

Where in-house wins

When the use case is small, internal, and unlikely to change, and when your engineers have the time and will keep the time. A read-only tool over one database is a genuinely good weekend project, and paying a vendor for it is paying for insurance you do not need.

Where we lose

The 24-month total

Two-year cost of a product server, four ways
Option	24-month total
Portkeel Product MCP	~$21,300
In-house build + self-maintenance	Build cost + your migration time
Contractor, rebuilt each revision	~$6,400 per rebuild

Decide in five questions

Do your engineers have two spare weeks this quarter?
Is the use case internal and unlikely to change?
Will someone own the next spec migration?
Does the server face customers?
Do you want protocol churn to be someone else's problem?

Three "yes" in the first three questions means build. A "yes" to either of the last two means buy.

Build vs buy questions

Who should build it in-house?: A team with spare senior engineering capacity, a simple internal use case, and someone who will own the 2026-07-28 migration and the next one. If all three are true, build it. You will spend two or three engineer-weeks and own the result, which is cheaper than any vendor for a tool you will rarely change.
Who should buy?: A team whose engineers are the bottleneck, whose server faces customers, or who does not want to own protocol migrations. The maintenance is the cost people forget: a server built in a weekend still needs someone awake when the spec revises.
What does "buy" actually cost over two years?: For a product server, roughly $4,500 up front and $700 a month, which is $21,300 across 24 months with migrations included. In-house is the loaded cost of the build plus every future migration you will do yourself, which is cheap until it is not.
Is there a wrong answer here?: Yes: buying a generic hosted server for a use case that needed a curated one, or building in-house and then not maintaining it. A server nobody maintains is worse than no server, whichever way you acquired it.

See what this costs