How to Start an Agency in 2026 (Even If It's Just You)

Starting an agency used to mean assembling a team before you'd earned a dollar: a designer, a developer, someone to manage the project, and enough runway to pay them while you figured out sales.

That math is gone. In 2026, one person with an AI delivery stack can sell, build, and ship the kind of work that used to take a studio. The websites, client portals, booking systems, and internal tools that small businesses pay a web design agency or digital agency for are now things you can deliver in days, alone.

The short version of how to start an agency, before we unpack it:

1. Pick a narrow niche, by client type or deliverable.
2. Productize one fixed-scope, fixed-price offer.
3. Connect Floot to Claude over MCP as your delivery stack.
4. Win your first clients by pitching working demos, not proposals.
5. Deliver in the open with live preview links, publish in one click.
6. Turn every project into a monthly retainer.

The stack this guide is built around is specific: Claude, connected to Floot over MCP. Your whole agency, from the first demo to the published site to the Tuesday-morning edit, runs out of a chat window. Here's each step in detail.

The one-person agency is now a real business

Here's what changed: what clients buy hasn't moved, but what it costs to deliver has collapsed. A salon still wants bookings. An accountant still wants a client portal. A gym still wants a dashboard. They don't care how many people are behind it, they care that it works and shows up on time.

AI collapsed the delivery side. Design, copy, code, and hosting used to be four skills and four line items. Now they're one conversation.

The honest caveat: the barrier dropped for everyone, so "I make websites" is a crowded sentence. The one-person agencies that win pick a niche and out-deliver on speed. Which brings us to step one.

Step 1: Pick your agency niche, not "everything"

The narrower your offer, the easier the sale. "I build booking sites for salons" closes faster than "full-service digital solutions," because the salon owner can immediately picture the thing they're buying.

Pick a lane by client type (realtors, coaches, clinics, restaurants) or by deliverable (booking sites, client portals, lead-gen sites, internal dashboards, AI automations). Ideally both. You can widen later; you can't easily un-blur a generic pitch. This works whether you're framing yourself as a web design agency, an AI automation agency, or just "the person who builds software for salons."

A bonus you'll feel by client three: a narrow lane means every project teaches you the next one. Your second salon site is faster than your first, and your fifth is a template with a new coat of paint.

Step 2: Productize your agency offer

Sell a package, not your hours. Fixed scope, fixed price, fixed timeline: "a booking site with payments, live in a week." Clients know exactly what they're getting, you know exactly what you're building, and nobody's counting hours.

Keep it to one flagship package and maybe one upsell. A menu of fifteen services reads as "we do anything," which reads as "we specialize in nothing."

Step 3: Connect Floot to Claude, because it's about to be your whole back office

Before the first pitch, set up the stack. In Claude's settings, add Floot as a connector; the Floot docs have the current walkthrough, and it takes about a minute. It works in ChatGPT and Cursor too, if that's where you live. (New to the protocol underneath? Here's a plain-English guide to MCP.)

From that point on, your Claude chat is an app platform. You describe what a client needs, and Claude builds it on Floot: frontend, backend, database, user accounts, file storage, payments, all of it, with a live preview link from the first edit and one-click publishing when it's done. No site builder, no plugin stack, no separate hosting account, no deploy step. If you want the full walkthrough of that build loop, we've written a step-by-step guide to building a website with Claude.

And because building over MCP uses zero AI credits on top of your Claude plan, the cost of building doesn't scale with how much you build. That detail quietly changes how you sell, which is the next step.

Step 4: Win the first three clients by building, not proposing

The old agency sales motion was a proposal, a deck, and a discovery call. You have a better move now: build the actual thing.

When a mockup costs a week, you pitch with slides. When a working site costs an afternoon in the same Claude chat you already have open, you pitch with the site. Describe the salon's booking flow to Claude, let it build on Floot, and walk into the conversation with a live preview link: "here's your booking site, it already works, want it?" You've skipped past every agency still scheduling discovery calls.

This is exactly why zero-credit building matters. Spec work used to be a real cost, so nobody did it. When a demo costs you nothing but an afternoon, the demo becomes your sales team.

Your first three clients are almost always warm: your network, local businesses you already frequent, communities you're already in. Build something real for each pitch. Worst case, you lose an afternoon. Best case, the demo is the deal.

Step 5: Deliver in the open, and let speed set your prices

Two Floot MCP habits do most of the heavy lifting on delivery.

First, share the live preview link with your client while you build. They watch the site take shape and give feedback mid-build, not at a big reveal at the end. "Actually, can the header be green?" costs you a sentence in the chat, not a revision cycle. Clients who feel involved sign off faster and argue less.

Second, when they say yes, publish from the same conversation. The site goes live at its own URL with one click, custom domain included, hosting that scales on its own. There's no handoff to a DevOps step, because there is no DevOps step.

Speed is your margin. When delivery takes days instead of months, a fixed price that felt expensive per-project becomes obviously cheap per-week for the client, and extremely healthy per-hour for you. Resist the urge to bill hours: hourly billing punishes you for being fast, and fast is your whole advantage.

Step 6: Turn every project into a retainer

The project gets you in the door. The retainer builds the agency into a business.

Every client who gets a site needs edits, seasonal updates, and, sooner than you'd think, more software: a portal for their customers, a little CRM, an automation for the thing they still do in spreadsheets. Because every one of those is a conversation in the same chat where you built their site, a monthly care-and-updates retainer is high margin for you and a no-brainer for them.

One person with a dozen retainers is a real business with real recurring revenue, and no payroll.

What it costs to start an agency this way

Close to nothing fixed. Your delivery stack is two flat plans: your Claude subscription and a Floot subscription that covers building, hosting, the database, and publishing. Floot MCP doesn't meter your tokens or sell credit packs, so demos, revisions, and retainer work never generate a usage bill. Your costs stay flat while your client count grows, which is precisely the opposite of how agency payroll works.

If you're weighing platforms for the delivery side, our comparison of the best AI app builders for non-technical founders covers the field.

FAQ

What is Floot MCP, exactly?

MCP is the open standard that lets Claude connect to outside tools, and Floot's connector puts a full app platform on the other end of it. Once connected, Claude can create projects, build features, and publish sites on Floot directly from your chat. Setup is a one-time thing in Claude's settings and takes about a minute.

Do I need to know how to code to start an agency?

No. Describing what the client needs in plain English is the interface, and non-technical founders are already building real apps this way. What you actually need is the ability to listen to a business owner and translate their problem into a clear ask, which is a rarer skill than coding anyway.

What should I charge?

Price the outcome, not your time. A booking site that fills a salon's calendar is worth a fixed, confident price, whatever it took you to build. Start where your niche's budget lives, and raise it every few projects until you feel resistance.

How many clients can one person actually handle?

More than the old model allowed, because delivery is no longer the bottleneck. Your real limits become sales and communication. Retainers help here too: a stable base of maintenance clients smooths the feast-and-famine cycle that kills most freelance practices.

What if a client needs something genuinely complex?

More is in-bounds than you'd expect: user accounts, payments, bookings, dashboards, client portals, and AI features are normal requests on Floot, not custom-development territory. If a project is truly beyond the platform, say so and refer it out; the trust usually comes back as more work in your lane.

Can clients own their site when we're done?

Yes. The site lives at its own URL with their domain on it. Keep it under your account as part of the retainer, or hand it off if the engagement ends. The retainer conversation is easier when they know they're not locked in.

The agency you can start this week

Pick a lane. Productize one offer. Connect Floot to Claude, build the actual demo for your first pitch instead of a proposal, deliver in days, publish with a click, and turn the project into a retainer.

The team you used to need is now a chat window: your judgment, Claude, and Floot on the other end of MCP. Connect them today, and you can be pitching tomorrow.