MCP vs calling the API yourself

These are not two levels of the same thing. The connector is a conversation and a script is a rule, and that difference decides which one suits a job.

What is identical

Almost all of the machinery, which is why the choice is not about capability.

• Both authenticate with the same API key.
• Both spend from the same credit balance at the same rates.
• Both run the same runnable tool set.
• Both refund a failed job in full.
• Both are subject to the same plan or credit pack requirement.

The connector is a client of the same endpoints in the Public API documentation. It does not have privileges a script lacks.

So nothing is unlocked by choosing one. What changes is who makes each decision.

That framing is worth holding on to, because the two look like beginner and advanced versions of the same thing and they are not.

A script is not more powerful. It is more repeatable, which is a different property entirely.

The connector suits undecided work

Reach for the MCP connector when the job is different every time and a person is in the loop anyway.

One photo that needs something doing, where what needs doing is a judgement call, is the ideal case.

It is also good when you are already in a conversation. Not switching windows is a real saving on small work.

And it needs no code at all, which puts it within reach of people who would never write a poll loop.

That reach is the real argument for it. Plenty of people who need one photo fixed will never open a terminal, and this is the route that works for them.

It also keeps the context. A follow up request does not need the previous one re explained, which a script has no equivalent of.

The cost is that every job needs a person present to ask for it. That does not scale and is not meant to.

There is a subtler cost too. An assistant interprets, and interpretation occasionally means running something slightly different from what you pictured.

On one photo that is cheap to correct. On fifty it is a problem you would rather have prevented.

A script suits decided work

Write the script when the decision is already made and the only thing left is repetition.

Four hundred identical cutouts is not a conversation. Nobody should be typing four hundred requests.

A script also runs without anyone present, which a conversation cannot. Work triggered by your own system belongs here.

You get exact control too. Retry behaviour, rate limit handling, and what happens on failure are all yours to decide rather than an assistant improvising.

The cost is setup time and a poll loop you have to write, plus errors you have to handle yourself.

Access is the other prerequisite people forget. A plan or a credit pack is needed either way, so neither route avoids that.

How to choose in one question

Ask whether the job will be identical next time.

That single question sorts almost every case, and it is more reliable than asking which one sounds more professional.

If yes, and it happens more than a handful of times, write the script. Repetition is what code is for.

If no, and a person has to decide each time anyway, use the connector. You are not automating a decision, you are removing a window switch.

The middle ground is real too. Plenty of work starts as conversations and turns into a script once the pattern becomes obvious.

Starting in the conversation is a reasonable way to find out what the script should do.

Most people who use both end up drawing that line naturally within a week.

The line usually lands somewhere around ten files. Below that the conversation is faster, above it the script is.

Whether the work has to run unattended shifts that line hard. Anything that must happen without a person is a script at any volume.

Cost is not a useful tiebreaker here, because both spend the same credits on the same jobs. Only your time differs.

Setup time is the thing you are really trading, and it is paid once for a script and never for the connector.

Weigh it against how many times the job will come back rather than how big it feels today.

One trap worth naming. An assistant can be asked to loop through several files, and it will try, but it is a slow and expensive way to do something a script does in one pass.

A batch is a script job, and what that looks like in practice is in bulk cutouts from a script I wrote.

One more thing worth saying plainly. Neither route lets an assistant spend credits you did not agree to, because every job still comes from a request you made.

Neither is more advanced. They answer different questions and most workflows want both.