The 7-Server Stack: How East Africa's Coordination Infrastructure Works Together

I've been building AI infrastructure for East Africa for the past 18 months. Seven MCP servers. Seven coordination failures. One underlying theory.

The Theory

Markets fail when participants lack the information to coordinate. In East Africa, the failures are concrete:

• A farmer sells maize at harvest when prices are lowest, not knowing Nairobi prices are 40% higher
• A small trader can't access credit because their M-PESA history isn't legible to banks
• A jua kali welder is paid below market because neither party knows the market rate
• A household goes uninsured against drought because parametric products aren't findable

Each of these is an information asymmetry problem. The technology to solve them exists. What's missing is the query layer.

The Stack

• mpesa-mcp → payment execution + transaction intelligence
• bima-mcp → insurance products + parametric risk scoring
• mkopo-mcp → alternative credit scoring from M-PESA patterns
• soko-mcp → commodity price intelligence across 8 markets
• sifa-mcp → portable reputation + skills passports
• kazi-mcp → labor market matching + wage benchmarking
• wapimaji-mcp → drought phase data across 47 counties

How They Compound

The power isn't any single tool - it's the combinations. An AI agent helping a farmer can simultaneously:

• Check current maize price in Nairobi vs local market (soko-mcp)
• Assess drought risk for their county (wapimaji-mcp)
• Find parametric crop insurance that pays out automatically (bima-mcp)
• Check if their M-PESA history qualifies them for input loans (mkopo-mcp)
• Verify their reputation score if selling through an aggregator (sifa-mcp)

That's five coordination problems solved in a single agent session. Previously, solving any one of them required navigating multiple institutions, phone calls, and days of waiting.

What's Next

The second layer is execution: tools that don't just provide information but actually trigger actions - STK Push payments, insurance policy enrollment, credit applications. Some of that already exists in mpesa-mcp. The rest is coming.

The goal isn't an app. It's infrastructure. Apps are built on top; they can be Swahili-native, SMS-delivered, USSD-based, or voice-first. The coordination layer stays constant.

All seven servers are live on PyPI. All are indexed on the Glama MCP directory. All are MIT-licensed.

pip install mpesa-mcp bima-mcp mkopo-mcp soko-mcp sifa-mcp kazi-mcp wapimaji-mcp