A two-person startup’s bet on MCP-native content
Waldium, a two-person, YC-backed startup founded by Amrutha Gujjar and CTO Shivam Singhal, runs an agentic CMS for businesses from a single Next.js deployment on Vercel. The platform automates content research and creation while giving each customer blog its own MCP server endpoint, which AI agents can query directly.
That MCP capability is the product’s defining feature, but it was never the original plan. Waldium initially built blogs for human readers. What shifted the focus was a pattern Amrutha kept observing: developers and technical teams were pulling blog content into coding environments, Claude Desktop, and ChatGPT, using published material as live context inside the tools where they worked. The browser tab was still part of the picture, but it wasn't the whole story anymore.
“The people consuming content are not even necessarily people these days. It's oftentimes agents consuming content.”
Amrutha Gujjar
That realization made MCP the right primitive. Each customer blog gets its own MCP server endpoint, so an AI assistant can search, retrieve, and interact with content without requiring a human to leave a workflow. A technical founder who doesn’t think of themselves as a marketer can create a blog post from the same place they’re building a feature, as Amrutha puts it.
From DIY infrastructure to a single deployment
Supporting hundreds of customer blogs with custom domains and unique, branded MCP servers, each with subdomain generation, SSL, and live install pages, is a serious infrastructure problem. Waldium originally assembled its own pipeline on AWS Amplify with GitHub Actions, manual deployment configs, and hand-managed SSL. As tenants grew, so did the maintenance burden. The team moved to Vercel early, and the developer experience was what stuck.
“The developer experience is so seamless,” Amrutha says. “It allows our team to focus on what we do best.”
Shivam points to the sustained simplicity of the mental model: “It's very simple now: push to a particular git branch, get a preview deployment, get a production deployment. It works cleanly with authentication, with our databases. It just works.” The integrated storage layer, with Neon and Upstash alongside the application in a single console, gave the team what he calls “a single pane of glass into all the parts of our system.”
Waldium’s MCP infrastructure specifically relies on Vercel for Platforms: a single Next.js application serves every customer blog, MCP endpoint, and custom domain. Vercel Middleware handles routing dynamically, sending each query from an AI agent to the right tenant. The Vercel Domains API provisions custom domains in seconds, with SSL certificates issued and renewed automatically. When the team evaluated MCP-specific hosting providers, they ran into subdomain caps in the dozens. Vercel handles tens of thousands.
“It allowed us to think at scale without having to worry about scale.”
Amrutha Gujjar
Onboarding now works in one flow: a customer signs up, gets a unique subdomain, and walks away with a sitemap, LLMs.txt, robots.txt, an MCP install page, and a live MCP endpoint, all in under five minutes.
Distribution without per-tenant work
Because every customer gets a dedicated MCP endpoint without per-tenant infrastructure overhead, customers gained a distribution channel that didn’t exist before. Their content is queryable inside AI assistants, not just discoverable through search.
Saphira AI, a safety compliance company publishing highly technical educational content, illustrates the shift. A standard content push previously meant weeks of manual research, with someone going through thousands of pages of ISO standards to identify what customers would find useful. Waldium replaced that with research agents that continuously build company and industry profiles, flag newsworthy updates, and generate content at a scale no traditional team could match. Saphira AI produced over 1,000 posts in a month, and research time dropped from weeks to hours.
Amrutha frames that capability as the new baseline for a business with technical credibility:
“Before, if you were starting a business, you had a landing page. Today, if you're starting a business, you have to have a corpus of data about your business that an LLM can consume.”
Amrutha Gujjar
Scale without the scale problem
The current state of the platform, from figures Waldium maintains about its own deployment, shows what the unified architecture buys them:
Waldium on Vercel
- 500+ customer blogs served from a single Vercel deployment
- New customers get MCP endpoints live in under 5 minutes
- AI query response times consistently under 50ms globally
- 45% lower infrastructure cost vs. per-customer deployment model
- One customer generated 1,000+ posts in a single month; research time cut from weeks to hours
Amrutha’s founding frustration, that infrastructure friction was always the part that slowed the product work, now reads as a design principle the company has validated. “Once you understand how good things can be,” she says, “it's really hard to go back to a product that hasn't given as much intentional thought toward good design.”
