The bottleneck just moved: why African startups should care about MCP – Disrupt Africa

The bottleneck just moved: why African startups should care about MCP - Disrupt Africa


Earlier this year, our four-person studio cleared a 126-item product backlog in a single 48-hour stretch. A few weeks later, an AI edited a 268,000-word manuscript overnight — coordinating a dozen sub-agents while the author watched from a terminal, writes Barrett Nash, former-founder of Rwandan startup SafeMotos and now CEO of InfiniteUp.

Neither happened because AI suddenly got smarter. Both happened because of a quiet piece of plumbing called MCP, and it deserves more attention from African founders than it’s getting.

What MCP actually is, in plain language

MCP — the Model Context Protocol — is an open standard, introduced by Anthropic in late 2024 and since adopted across the industry, that lets AI assistants operate software rather than just talk about it.

Think of any piece of software as having a front desk: a published list of requests it will accept. MCP lets an AI walk up to those front desks itself. Before MCP, you described a problem and the AI described a solution — then you did the work. With MCP, it does the work. Type “make the button on the contact page green,” and in your website builder, the button turns green.

That’s the entire idea. Everything interesting is downstream of it.

Why this matters more in Africa than almost anywhere

African tech has a long habit of turning constraints into architecture. M-Pesa didn’t wait for branch banking; it leapfrogged it. The constraint MCP erases is headcount.

A four-person team that restructures its work around these tools can genuinely outproduce a much larger one that has bolted a chat window onto an unchanged process. We know because we are one: 40+ production apps, users in 117 countries, including DOER Business, a product we built specifically for African small businesses. Meanwhile, the cost side has collapsed — our own AI bills fell roughly 100× in three years. Intelligence stopped being a premium feature and became affordable infrastructure, priced for markets where every dollar of runway counts.

For startups that could never afford a twenty-person engineering floor, that combination — AI that does the work, at prices that keep falling — is not an efficiency gain. It’s a different game.

What it looks like in practice

Real examples from our own workshop this year:

  • Operations that run themselves. Connect an AI to your inventory, your payment provider and WhatsApp, and “check stock, reorder, notify the customer” stops being someone’s job and becomes a sentence.
  • Localisation as market-building. Asked to translate our site for Gulf markets, the AI didn’t translate it — it deployed sub-agents that rebuilt every page’s argument for a different audience. The same approach applies to Swahili, Amharic or Lingala: not your words in another language, but your business argued to a different room.
  • Grounded strategy. We fed our raw search data straight into the model and let evidence, not opinion, set our content strategy. Organic traffic responded within weeks.
  • Finishing what AI started. A wave of founders have “vibe-coded” apps to 80% in tools like Lovable and Bolt — then stalled at the last, hardest 20%. Rescuing those apps is now real engineering work, and the demand is continental.

The honest caution

MCP hands AI a pair of hands, not judgment. We once watched an agent of ours report five consecutive rounds of “improving results” while measuring nothing at all — the numbers were real, the link to reality had quietly broken. Our rule since: more AI throughput means more human verification, not less. The teams that win with this won’t be the ones that automate everything; they’ll be the ones that know exactly what to supervise.

Where to start

You don’t need a research budget. Pick one painful, well-documented workflow — documentation quality is the real fuel here — connect an AI to the systems it touches, and supervise the first runs closely. If you want a shortcut: we’re offering a free 30-minute MCP consultation to any company that wants a concrete map of what this could do for their business — no pitch, just the map. The tools are the easy part. The reorganization is the work, and it’s the most interesting work in software right now.

Barrett Nash is the CEO of InfiniteUp, a four-person software studio in Nanaimo, Canada that has shipped 40+ apps to users in 117 countries — including DOER Business, built for African small businesses, and HearWB, the National Acoustic Laboratories’ Team of the Year for 2024. He previously co-founded SafeMotos in Kigali, Rwanda. African companies can book a free 30-minute AI consultation here.



Source link

Leave a Reply