A couple of months ago, I wrote about holding two truths in the same hand. Real concern about what AI is doing to creative value, weighed against usefulness of a tool that can synthesise the mess of a working brain. I landed on cautious optimism in the valley of despair, and a promise that whatever we built wouldn't be a bolt-on AI for AI's sake. (Missed it? Read it here).
Since then, we've been in the weeds, building our MCP server brick by brick, tool by tool. Not repurposing our old API and slapping "AI-ready" on it, but building a brand new one, designed from the ground up for assistants like Claude and ChatGPT to use as efficiently as possible. It's in open beta now. And it feels a little bit like magic. Let’s run through what it does, where it shines, and where you should keep your hands on the wheel, because magic isn’t always right.
What it actually is
MCP stands for Model Context Protocol, an open standard for letting AI assistants talk to outside tools. Our connector is Streamtime's MCP server. Once you connect it inside your assistant (the setup lives on your assistant's side, not inside Streamtime), you can talk to your account in plain language. "What jobs are in play?" "Add a design item to the Womens World Cup Merch job." "Who’s free to work on the Coffee Rebrand next week?" The assistant works out which Streamtime actions or searches it’s able to help with.
You’re giving your assistant exactly your permissions, no more, no less. If you can't see job pricing in the app, neither can it. Connecting it is like handing someone the keys to your car, but you’ll want to keep your hand on the steering wheel (or hovering over the handbrake, like a nervous parent watching their sixteen year old).
Why MCP first?
So yes, we could have slapped in an AI agent on top of everything, ridden the wave, put AI in for the sake of AI. But to us, AI isn’t a feature, it’s a method: a way to solve a problem. And the problem should always come first, not the solution. Prioritising the MCP gives our customers the keys to the castle and learning what they want to do with those keys before promoting some of the workflows in Streamtime. It also allows customers to connect ST data with any other tool they might have in their kit, which an in-app agent doesn’t do.
That said, we are working on ways to integrate AI into the product, based on what we’ve heard. It’ll only get better from here.
What does it do?
When using the MCP, the safest and most common use case is simply reading your data. Ask it to find Jobs, Items, To Dos, Companies and Contacts, using your own language to build filters. Check who on your team has capacity across a date range. List To Dos, scheduled and logged, for a given week, check your Priorities Boards, report on your Jobs. In our current cohort of beta customers, 57% of all actions are read actions: searching and reporting.
Then there's building. The Job is the centre of everything in Streamtime, and you can now create one multiple in ways, much more quickly. From scratch, from an uploaded CSV or brief, or by duplicating something similar you've done before, all through your AI tool. One big time saver for people using MCP is setting item dates. For accurate reporting, item dates are king, and now you can clean up your setup hygiene after the fact. Try setting or adjusting item dates based on logged To Dos, or creating To Dos based on item dates and team availability. Around 27%* of actions are write actions, most commonly setting Item dates and updating To Dos in bulk.
Creating Expenses, Quotes, Invoices and Purchase Orders is also possible, with everything created by MCP landing as draft, so you review and send it from within Streamtime. For some actions, your AI may ask you a clarifying question before it creates anything. Your hands firmly in charge.
Where to be careful
Never forget, you’re still trusting the machine: your client is interpreting your plain language, so review anything that matters, especially money stuff and deletions. Deletions are permanent, just like in Streamtime.
Spend wisely: don't spend your tokens on things Streamtime already does well. Batch edits from a list view and CSV exports are still faster straight out of the app. Moving To Dos around the Schedule one by one is a quick and visual way of shuffling things. The MCP earns its keep on the heavy lifts like big job setups, bulk date-setting and, more than anything, when you point it at your Streamtime data alongside something else.
I'll leave you where I keep landing. The blank page problem hasn't gone anywhere. The human still has to start, think, bring judgement, leave their fingerprints. But give yourself permission to surrender the effort it takes to build the tedious scaffolding around the work you want to spend time on.
*For those of you asking what the last 16% of actions are? 10% are setup calls, providing your model some information and context, and 4% shared among less common tool calls: getting user availability, who am I, updating job details, among others.
COMING SOON: We’re in the middle of a rebrand rollout to our product. Watch this space for sneak peeks!
.png)
