Ask someone on your team to show you the last three things they wrote this week — not the polished ones, the working copy. A product description drafted in ChatGPT, a support reply tidied up in Claude, a headline knocked out in Cursor. Read them side by side and there's a fair chance they sound like three different companies. Same person, same brand, same afternoon.

This is the brand consistency problem nobody put on a roadmap, and it's worth being honest about where it came from. AI didn't create it. AI industrialised something that was already broken, and the broken thing is the bit marketing has always been least keen to examine.

Every business of any size has a brand guide somewhere. A tone-of-voice document, a page of do's and don'ts, a glossary of the words you use and the ones you'd never be caught dead using. In most businesses that document lives in a Notion page, or a PDF on a shared drive, that precisely nobody opens before they start typing. That was true long before anyone had heard of a large language model. The guide existed to be admired rather than used, a piece of brand theatre that made everyone feel professional and changed nothing about the output. I've written before about marketing's habit of impressing itself rather than doing the work, and the unread style guide is that habit in its purest form.

What AI did was take that tolerable, low-level inconsistency and put it on a production line. When one copywriter ignored the guide, you got one off-brand paragraph and a mild wince. Now everyone on the team has half a dozen AI tools, each of them a fresh mouth with no sense of your brand, all generating copy at a volume no human team could match. The guide is still sitting in Notion, still unopened, and now it's being ignored six times faster.

The fix that finally targets the right layer

Which is where a tool like Brivvy comes in. You define your tone, audience, glossary and rules once, and it pipes them straight into Claude, ChatGPT, Cursor and whatever else your team writes in, so that when someone generates copy in a connected tool the voice is already applied. There are sliders for warmth, confidence and formality, a preferred-and-avoid word list, the works. Underneath, it's a brand-voice MCP: a single voice config that every connected tool reads from through the Model Context Protocol.

The architecture is what makes this interesting, not the feature list. For twenty years the answer to brand inconsistency was to write a better document and nag people into reading it. Brivvy's bet is that this was always the wrong layer. You don't fix consistency by improving the guide nobody opens, you fix it by moving the rules to where the writing actually happens, so following them becomes the path of least resistance rather than an act of willpower. That's the right instinct. Put the voice in the plumbing and the plumbing carries it, whether or not anyone remembers the guide exists.

And the timing is real rather than hype. In under two years MCP has gone from an Anthropic side-project to shared plumbing that OpenAI, Google and Microsoft all support, which means piping one voice config into every AI client your team uses is no longer an exotic bet. It's the kind of thing you can stand up quickly and reasonably expect to still work next year.

A protocol can't give you a voice

The catch is the sort of thing launch copy tends to skip. A protocol can distribute a voice, but it can't hand you one worth distributing.

If your brand voice is genuinely worked out — if someone has done the slow, unglamorous thinking about who you're talking to and how you want to land with them — then piping it into every tool is a clean win. But that isn't the state most businesses are in. Most have the unopened Notion page. I'm not sure many teams can even tell the difference between a voice they've genuinely decided on and one they've merely described, and that uncertainty is part of the trouble. Wire a vague, half-considered voice into the plumbing and you don't get consistency, you get your mediocrity enforced across six tools with perfect fidelity. The slider reads "confident, warmth 60%" because someone dragged it there on a Tuesday, and now that guess is your house style everywhere, applied automatically, until someone thinks to change it.

So the tool solves distribution while leaving definition exactly where it was — on you. And definition is the hard, human part marketing has spent years avoiding, because it can't be bought, awarded, or knocked out in an afternoon. Brivvy will cheerfully enforce a brand voice you haven't actually decided on. That's less a knock on the tool than a warning about the order of operations: sort the voice, then automate it. Automate it first and all you've done is industrialise the guesswork.

Someone has to own the voice

There's a second thing worth flagging, and it's less about marketing than about how your stack quietly assembles itself. A brand-voice MCP is one more component wired into every AI client your team uses, an asset and a liability in the same breath. I've written about agent sprawl and the question of what each tool is allowed to touch: the discipline of deciding, deliberately, who owns each moving part before you've got forty of them. A shared voice config is exactly that sort of moving part, and it comes down to three plain decisions, none of which need a platform to make. Write down what your voice actually is before you wire anything up. Name the one person who owns the config and signs off changes to it, because a glossary edit now reshapes every output across the business. And put a date in the diary to revisit the sliders, rather than letting a Tuesday-afternoon guess ossify into house style. Do those three and the shared config is genuine leverage. Leave it to assemble itself and it's just a faster way to be wrong in unison.

The plumbing has finally caught up with the problem, and that's good news: for the first time, keeping everyone on-brand is a solvable engineering question rather than a permanent low-grade nag. But it only pays off if there's something worth pushing through the pipes. So before you connect anything, do the part no protocol can do for you and work out what your business actually sounds like, and why. Get that right and the tools will carry it everywhere you write. Get it wrong and they'll do precisely the same — which is rather the point.