Should You Build an MCP Server Before a Dashboard? A Founder's Post-Mortem

Samuel Afriyie's Indie Hackers post from April caught my attention — "I should have built the MCP server first, not the dashboard." We made the same mistake. 1ClickReport's dashboard shipped first; the MCP server came second. Six months in, the MCP layer is the part of the product that's actually winning. Here's the honest post-mortem on what we'd do differently if we were starting today, with the numbers that drove the realisation.

The mistake we made

1ClickReport launched in late 2025 with the standard SaaS architecture: a hosted dashboard, OAuth-connected data sources, a pricing page, a free trial. By any conventional SaaS playbook standard, this was correct. We were building a marketing analytics tool, and marketing analytics tools have dashboards.

What we underestimated: the dashboard SaaS market is brutally crowded. Supermetrics, AgencyAnalytics, Databox, Whatagraph, DashThis, Funnel.io — all dashboard tools, all with bigger marketing budgets and longer install bases. Differentiating on "we have a nicer interface" was always going to be a slog.

We added the MCP server about 5 months in, almost as a side project — "let's expose our data layer to Claude users." That side project is now the part of the product driving the highest-quality signups.

The numbers that changed our mind

Three numbers, all from our own GA4 + signup data over the last 90 days:

Conversion rate of MCP-discovered users is ~3x dashboard-discovered users. People who find us through Claude's MCP directory, Smithery, GitHub MCP lists, or through "best MCP for marketing" content are pre-qualified — they already know what MCP is, they have a specific use case, they convert.
Cost per signup from MCP channels is roughly 80% lower. Listing 1ClickReport on the Anthropic MCP directory cost us 30 minutes and zero dollars. The signups from that listing — and from organic discovery of our GitHub MCP toolkit repo — have functionally no acquisition cost.
Churn is lower for MCP-first users. Users who first discover us through their existing AI workflow (Claude Desktop, Claude Code) tend to integrate us into a daily habit faster than dashboard users, who have to remember to log in.

The dashboard isn't dead. It serves a real audience — non-technical marketers who want a polished UI. But the MCP channel is generating better unit economics with less marketing effort.

Why MCP-first works structurally better in 2026

1. Distribution is built into the AI workflow

A dashboard tool needs you to acquire users, get them to your site, convince them to sign up, and get them to log in regularly. An MCP server gets discovered inside the AI tools your users already live in. The distribution is the workflow.

2. Standards mean interoperability for free

Build a Claude integration in 2024 and you only worked with Claude. Build an MCP server in 2026 and you work with Claude, ChatGPT, Cursor, Windsurf, Continue, and whatever new MCP-compatible client launches next. The standard pays compounding dividends.

3. The data layer is the durable moat, not the UI

SaaS dashboards are increasingly commoditised — Looker Studio, Looker, Tableau, Metabase all compete on price and features. What's hard is the data integration layer: OAuth connections, rate-limit handling, schema normalisation, real-time fetching. If you build that as an MCP server, you can plug it into any UI later. If you build it as a UI-first product, you're stuck with your UI.

4. AI-native users have higher ceiling

Users who live in Claude Code, ChatGPT, or Cursor tend to be more technical, more willing to pay for productivity tools, and stickier once they integrate something into their workflow. They're a smaller TAM than "all marketers" but a higher conversion-and-retention TAM.

What we'd do differently if starting today

Months 1-3: Build the MCP server only

No dashboard. No pricing page. Just an OAuth-based MCP server that exposes your data via well-named, well-documented tools. Hosted at mcp.yourbrand.com. List on Anthropic's directory, Smithery, Glama, GitHub MCP lists. Documentation in a public GitHub repo. Free tier.

Month 4: Add a minimal dashboard

Once you have MCP users, build a thin dashboard for the connection management + billing flow. Don't try to compete with Looker Studio on visualisation. Make the dashboard a control panel, not the product.

Months 5+: Layer in dashboards or workflow tools

Now you can decide whether to invest in dashboard depth or in higher-order workflow tools (alerts, scheduled reports, action automation). Both are options. Both should be optional layers on top of the core MCP — not the core itself.

Who shouldn't do MCP-first

Some categories aren't right for this architecture (yet):

Consumer products. Your end users probably don't use Claude or Cursor. MCP-first only works when your buyer is technically inclined enough to install MCP servers — which is still a B2B-skewed audience in 2026.
Highly visual workflows. If your product's core value is heavy data visualisation (e.g., financial charting, design tools, video editing), a dashboard IS the product. MCP can be supplementary.
Compliance-heavy verticals. Healthcare, banking, government — these buyers don't yet have institutional comfort with AI-integrated workflows. MCP-first signups would be near zero. Build the conventional product first.
Products with strong network effects in the UI. Marketplaces, social tools, collaboration platforms — the UI IS the network. MCP can extend but not replace.

For B2B SaaS where the buyer is technical and the value is data-integration / analysis / automation: MCP-first is the better default in 2026.

The honest tradeoffs

MCP-first has costs we don't want to gloss over:

Smaller TAM in 2026. The pool of "people who use Claude Desktop and want marketing analytics in it" is genuinely smaller than "all marketing teams." It's growing fast but it's not where the majority is today.
Investor narrative is harder. "We're an MCP server" is harder to pitch to VCs in 2026 than "we're a marketing analytics dashboard." Investors don't yet know what MCP is. Founders need to explain.
Security scrutiny is higher. MCP servers are still earning institutional trust. Post-CVE-2026-30623, every MCP gets a security audit from technical buyers. You need to be ready for the conversation.
Pricing models are unclear. Per-seat? Per-query? Per-connector? The pricing playbook for MCP servers isn't established. You'll iterate more than you would on a dashboard product.

These are real costs. But the structural advantages of MCP-first — distribution built into the workflow, standards-based interoperability, AI-native user base — outweigh them for the right product category.

Frequently Asked Questions

Do MCP-first companies actually exist?

Yes — and the count is growing fast. As of May 2026, the Anthropic MCP directory lists hundreds of MCP servers, and Glama tracks roughly 23,000+ servers including community-built ones. Many are open-source side projects, but a growing number are SaaS companies with MCP as the primary distribution surface.

Isn't building an MCP server harder than building a dashboard?

About the same complexity, different shape. An MCP server requires defining clear tool boundaries, handling auth flows, and supporting multiple MCP clients. A dashboard requires UI work, state management, and visualisation. Both involve real engineering — the question is which surface attracts your target customer.

Can I add an MCP server to an existing dashboard product later?

Yes, and many companies are doing this now. The retrofit is harder than building MCP-first because you've already shaped your data layer around dashboard needs. But it's very doable — we did it, Supermetrics is doing it, Databox is doing it. Expect 2-4 months of engineering.

How do MCP-first companies make money?

Same models as any SaaS — subscriptions, usage-based pricing, per-seat for teams. The difference is where customers discover and adopt you, not how you charge. Our pricing is straightforward $25/mo (Pro) and $99/mo (Premium); the MCP-first acquisition doesn't change the unit economics meaningfully.

Is the dashboard market really dying?

Not dying — but increasingly commoditised. Looker Studio is free and capable. The premium tier of dashboard tools (Funnel.io, Tableau) is alive and well. The middle tier ("slightly better than Looker, less powerful than Tableau") is where MCP-first products are eating share faster.

What about products that need both MCP and a polished dashboard?

That's our position now and it works fine — you just need to be honest about which surface drives growth. For us, MCP discovery converts at 3x dashboard discovery, so MCP is the primary acquisition channel and the dashboard is the secondary surface for non-technical buyers. Both matter; sequence them right.