Picture this: A three-person marketing team launches a product across Meta, Google, and TikTok. They're using Claude to write copy, ChatGPT to brainstorm strategy, Cursor to tweak landing page code. Each AI generates creative assets. None of them talk to each other. By week two, the brand looks like three different companies.
It's a small problem that scales badly. And Ray Fitzgerald thinks he's found an elegant, if not entirely obvious, solution.
Fitzgerald's company is called Bloom—a two-person operation that emerged from Y Combinator and went live earlier this year with something unusual: a Model Context Protocol integration. The technical details matter less than the premise. Bloom wants to become the single source of truth for your brand, accessible to every AI assistant you use. One brand model. Many agents. No more uploading your logo to six different tools or re-explaining your color palette in every chat window.
"Every agent calls the same canonical source of truth," the company's homepage announces, with the kind of certainty that only a startup can muster.
That source? Bloom's brand ingestion engine, which the company claims can extract a complete visual identity—logo, typefaces, color schemes, composition patterns, the whole aesthetic footprint—from a website or Instagram profile in about 30 to 60 seconds. Upload once, supposedly, and you're done.
A Protocol Play in a Crowded Market
The mechanics are straightforward, at least on paper. Paste Bloom's MCP URL into Claude Desktop, ChatGPT, Cursor, VS Code—anywhere that speaks the protocol. OAuth 2.1 handles the sign-in. No wrestling with command lines or rotating API keys, which is a relief for anyone who's ever maintained developer credentials across a dozen platforms.
Once connected, the AI can generate on-brand images: ads, social posts, email headers, website banners. It resizes them across aspect ratios—square, portrait, landscape, the usual suspects—and can batch up to 30 images in a single conversation. You feed it product shots, competitor ads, moodboards. Edits happen through natural-language prompts. The system renders galleries inline.
There's also a governance layer, because of course there is. Tool-level permissions, audit logs, allow/deny lists, a kill switch for when your autonomous marketing agent starts generating truly bizarre creative at 3 a.m. It's all documented under something called "Bloom Technologies," though the exact boundaries of that entity remain a bit murky.
Recent developer guides—published over the past few weeks—sketch out three core workflows. One involves an AI marketing agent that monitors click-through rates, spots creative fatigue, and autonomously churns out new variants. Another scripts an entire product launch from a single prompt. The third tackles ad creative at scale for Meta, Google, TikTok, LinkedIn, Pinterest. The usual suspects.
The pitch, essentially, is that Bloom becomes infrastructure. A brand-aware layer sitting beneath the chaos of competing AI clients.
The Proliferation Problem
Fitzgerald's timing reflects a real shift. The number of AI assistants embedded in developer and marketing workflows has exploded recently. ChatGPT, Claude, Cursor, lesser-known tools like OpenClaw, n8n, Windsurf. Each one can design a Facebook ad or an Instagram story. Each one starts from scratch unless something connects them.
The Model Context Protocol—Anthropic's attempt at a shared standard for AI tool integration—offers a potential answer. If every agent speaks MCP, they can theoretically tap the same brand model without custom integrations for each platform.
Bloom isn't alone in this realization. Canva introduced an MCP server exposing design tools and brand libraries. Typeface, aiming at enterprise customers, launched its own MCP with governance features on March 30. Layer rolled out what it calls a "Creative OS inside every AI you use" the same day—either a coincidence or a coordinated wave, depending on your level of cynicism. Skai followed on April 9 with MCP for its commerce media platform. Even non-design players like Third Bridge (finance research) and Blend (a lending platform) jumped in during the spring.
A pattern emerges: brand assets and rules live in specialized services. Agents call them via MCP. Companies stop re-uploading the same guidelines to a dozen different tools. Whether that vision actually materializes is another question.
Founder, Traction, and the Usual Caveats
Fitzgerald previously built Lucid, an AI-native word processor that's now shuttered. Before that, he spent nearly two years as a senior engineer at Inscribe, a Y Combinator alum from the Summer 2018 batch, working on AI agents for fraud review.
In an April LinkedIn post, Fitzgerald claimed over 40,000 people had used Bloom since launch and that revenue had tripled in the past month—figures that remain unverified beyond his self-reported update. The Y Combinator directory mentions a "viral launch" in December that pulled 15,000 signups and millions of impressions.
According to Bloom's careers page, the company has raised $1 million, with Guillermo Rauch—founder of Vercel—among the investors, though no independent press coverage or SEC filings corroborate these claims as of mid-May.
Pricing appears to be in flux. The current pricing page lists a Scale plan at $90 per month for 500 credits, with MCP and API access included across all tiers. Other sections of the site still reference a $20-per-month Plus plan, suggesting either a recent pivot or ongoing experimentation. Team features include shared workspaces, pooled credits, unlimited members, unified billing—the standard SaaS playbook.
Infrastructure or Feature?
Bloom frames itself as a "Brand OS," which is ambitious phrasing for a two-person company. The emphasis isn't on one-off asset creation but on powering workflows: agents that refresh creative when performance slumps, product launches scripted end-to-end, ad variants generated and resized on the fly.
Whether that resonates depends on several variables outside Fitzgerald's control. How quickly do marketing teams actually adopt agentic workflows? Will they trust a startup to hold their "canonical source of truth" when Canva, Adobe, and half a dozen enterprise platforms are building comparable MCP connectors?
For now, Bloom's bet is that the proliferation of AI clients transforms brand consistency from a manual governance headache into a protocol problem. And protocols, in theory at least, offer neutral ground.
Whether a two-person team can hold that ground is the real test. The infrastructure play is elegant. Execution at scale? That's a different story entirely.
