PayLink eCommerce Financial Assistant

A WhatsApp agent showcasing PayLink, my open-source payment framework, integrated with Sonar’s models to enable financial insight and financial advice

Comment

Inspiration

This project began as part of a larger vision.

I was building SmartSoko, an AI-powered shopping assistant to transform how people shop in Kenya using artificial intelligence. After completing the assistant and cart functionality, the next logical step was payments — and naturally, I turned to M-Pesa, Kenya’s leading provider.

But integrating M-Pesa wasn’t smooth. The APIs weren’t developer-friendly, especially for those building AI agents. Every step felt like reinventing the wheel.

Around that time, I discovered Anthropic’s Model Context Protocol (MCP) — a protocol designed to let tools interact with AI agents. At first, it didn’t click. But as I kept wrestling with converting the M-Pesa API into a usable tool, it hit me:

Every AI developer in Africa will run into this same problem — manually adapting payment APIs into formats agents can understand.

That realization shifted my focus.

Instead of solving the problem just for SmartSoko, what if I solved it for everyone?

Around that same period, I saw Perplexity AI launch Perplexity Shop and Amazon roll out their shopping assistant. These moves validated what I was doing with SmartSoko. I looked deeper into how these companies handled payments — often through AI-friendly tools like Stripe’s virtual cards.

That’s when it really sunk in: African developers didn’t have access to similar AI-compatible payment infrastructure.

So I circled back to MCP, this time with a bigger mission — to build a server that wraps payment APIs into agent-friendly tools.

I built the first version, DarajaMCP, and shared it online. It got a lot of attention. I even tagged M-Pesa on LinkedIn, hoping to spark collaboration.

But working with the M-Pesa sandbox revealed limitations: not all functionality worked, and production credentials were difficult to obtain. I started exploring other providers — Airtel, Stripe, Chipper Cash — and a new realization emerged:

This isn’t just about M-Pesa. It’s about building a modular, AI-native framework that supports all payment providers.

And that’s how PayLink was born — an open-source MCP server designed to wrap local and global payment APIs into AI-compatible tools. Built to power the next generation of intelligent agents, financial automation, and AI-driven commerce, PayLink is for developers across Africa and beyond.

What began as a personal pain point evolved into a platform — a foundation for financial access, developer productivity, and AI-first innovation in emerging markets.

What It Does

PayLink (Framework)

  • A standardized, MCP-compatible server that wraps payment APIs like M-Pesa, Airtel, and Stripe.
  • Enables AI agents to process payments, monitor transactions, and interact with financial systems seamlessly.

PayLink eCommerce Financial Assistant (Demo App)

  • Built on top of PayLink using Perplexity’s Sonar API.
  • Designed for merchants and online sellers to:
    • Search for real-time competitor pricing.
    • Analyze internal sales and transaction data.
    • Process payments using the PayLink SDK.
    • Get smart insights on pricing, inventory, and growth opportunities.

It’s like Perplexity + Mobile Money, tailored for the African context.

How I Built It

PayLink (Framework)

  • Built with FastAPI.
  • Implements MCP tools like stk_push to initiate payments.
  • Uses MongoDB to trace and persist transaction histories.
  • Exposes a developer-friendly interface for any AI agent.

PayLink eCommerce Financial Assistant

  • Integrated Sonar API for real-time web search and product insights.
  • Designed to work with both LLMs that support tool use and those that don’t.
  • Uses a modular architecture to allow smooth expansion to other providers.

Challenges We Ran Into

  • Sonar API does not support native tool calling — I had to parse and structure results manually.
  • M-Pesa sandbox unreliability and hurdles in accessing production credentials.
  • Ensuring agents remain stateless yet traceable during payment sessions.

Accomplishments We’re Proud Of

  • Created a functioning MCP-compatible SDK that works with and without native tool-calling LLMs.
  • Abstracted mobile money systems into a plug-and-play interface for AI builders.
  • Laid the foundation for financial inclusion infrastructure tailored for African developers.

What We Learned

  • Africa’s fintech infrastructure is rich in potential — but still lacks standardization.
  • MCP has the potential to become the bridge between agents and real-world actions.
  • Designing AI agents that handle payments requires a deep focus on safety, traceability, and data flow integrity.

What’s Next

  • Integrate more payment providers: Stripe, Airtel, Flutterwave, MTN MoMo.
  • Add dashboards: Transaction insights, sales charts, and external market data.
  • Build a WhatsApp wrapper: Empower small business owners with AI-driven financial advice — without needing a full app.
  • Collaborate with African fintechs: Push for open APIs and more developer-friendly access to production environments.

Built With

Share this project:

Updates