We recently shared how we added an MCP server on top of our Buy Now Pay Later API to make it agent-ready. Read the full post: https://tinyurl.com/4b4fxyvm.
That was the foundation. Now we’ve built the agent itself.
Our BNPL Agent sits on top of that MCP layer and gives users and other systems a conversational, autonomous way to interact with installment data, repayment plans, and transaction history.
How Its’s Built
We followed Anthropic’s guidance on building effective agents. The architecture is deliberately simple. No over-engineered orchestration. The agent uses our MCP server to discover and call BNPL capabilities, and exposes two clean interfaces:
- An agent endpoint for open-ended, conversational interaction
- A workflow endpoint for structured, deterministic tasks where precision matters more than flexibility
The agent supports the A2A (Agent-to-Agent) protocol, meaning it can be discovered and called by other agents. This is where things get really interesting.
What you can ask it?
The agent can retrieve transactions, generate repayment plans, simulate installment splits, and reason across all of it. Real questions it can answer:
- “How much interest did I pay last year because I split transactions into installments?”
- “What will my total installment payments be next month?”
- “Which of my recent transactions can I still split into installments without exceeding €150 per month in total repayments?”
- “Show me all my active repayment plans and when each one ends.”
- “If I split this €900 purchase into 6 installments, what’s the total cost including fees?”
These aren’t canned reports. The agent reasons across live data to compose answers.
The Real Power: Agent-to-agent Composition
A single agent answering BNPL questions is useful. But the real unlock is when our agent collaborates with agents we didn’t build.
Because it supports A2A, other agents can discover and call it as part of larger workflows. Some examples we find compelling:
- A personal finance agent calls our BNPL agent and an inflation data agent to answer: “Compare the fees I paid on installments this year with inflation. Am I better off having split those purchases or would I have saved money paying upfront?“
- A budgeting agent combines our repayment schedule data with income and expense data from a banking agent to say: “You can afford to split two more transactions this month without going negative.“
- A shopping assistant agent checks our agent for existing installment commitments before recommending: “You’re already paying €200/month in installments. Adding this purchase would push you to €280. Here’s a longer plan that keeps you under budget.“
- A tax advisory agent queries our agent for total interest paid across the year and factors it into a broader financial summary.
None of these require us to build integrations. The agents find each other, negotiate capabilities, and compose answers autonomously.
Why This Matters
APIs gave developers access to BNPL. The MCP server gave agents access. Now the BNPL Agent gives agents a peer they can collaborate with.
We’re moving from “integrate our API” to “let your agent talk to ours.” That’s a fundamentally different model for how financial services get consumed.
The infrastructure is ready. The interesting part is seeing what gets built on top.