First in East Africa on All Three AI Agent Protocols: MCP, A2A, and Google ADK

In 2024-2025, three significant AI agent protocols emerged:

1. MCP (Model Context Protocol) — Anthropic's open standard for tools and data
2. A2A (Agent-to-Agent) — cross-vendor agent communication protocol
3. Google ADK (Agent Development Kit) — production agent pipeline framework

Most developers picked one. Most regions had zero implementations of any.

I built all three — for East Africa — and I'm the first documented engineer in the region to do so.

Why All Three?

The ecosystem will consolidate, and first-mover implementations are valuable regardless of which wins. Being the reference for East Africa on all three means the region's infrastructure isn't locked to any single vendor's roadmap.

But practically: each genuinely solves a different problem.

MCP — For Wrapping Existing APIs

MCP is the right choice when wrapping APIs as tool calls an AI agent can invoke directly.

For East Africa:

pip install mpesa-mcp       # M-PESA Daraja API
pip install wapimaji-mcp    # Kenya water infrastructure
pip install swahili-health-mcp  # DHIS2 health data

# Add to Claude
claude mcp add mpesa -- mpesa-mcp
claude mcp add water -- wapimaji-mcp

MCP's value: reducing integration friction per developer to near-zero. The institutional knowledge lives in the MCP server, not in every app built on top.

Result: mpesa-mcp — v0.1.9 on PyPI, 400+ downloads, 12 countries, first African payment API in the MCP ecosystem.

A2A — For Multi-Agent Coordination

A2A is the right choice when multiple specialized agents need to coordinate on complex tasks — negotiating, delegating, sharing context.

For Kenya's civic stack, this means budget accountability that needs a financial agent + legal agent + Swahili summarization agent working together. kenya-a2a handles the message routing and capability advertisement.

Google ADK — For Production Pipelines

ADK is the right choice for production-grade pipelines with evaluation and observability. kenya-adk runs multi-step Swahili advisory workflows with integrated quality evaluation.

The Swahili-First Architecture

All three implementations share one principle: tool descriptions are written in Kiswahili.

When an AI agent receives a Swahili instruction and selects tools, it matches against descriptions. English descriptions introduce a translation step — a failure mode. Swahili descriptions eliminate it:

{
    "name": "mpesa_stk_push",
    "description": "Anzisha malipo ya Lipa Na M-PESA. "
                   "Tumia wakati mtumiaji anataka kulipa kwa M-PESA.",
}

Status

All MIT licensed. All part of a 110+ tool East Africa portfolio.

The Kenya MCP Hub is a CLI registry for all servers:

pip install kenya-mcp-hub
kenya-mcp-hub list

Portfolio: gabrielmahia.github.io