Build the next generation of European payments with us
import { experimental_createMCPClient as createMCPClient } from "ai"; const autopay = await createMCPClient({ transport: { type: "sse", url: "https://mcp.autopay.pl" },}); const { create_payment } = await autopay.tools(); const payment = await create_payment.execute({ amount: 129.99, currency: "PLN", method: "blik",}); // mocked response// -> { status: "authorized", paymentId: "pay_8f2c1d" }Autopay exposes an MCP server, so your agent can call it like any other tool.
autopaylab is Autopay's R&D program for merchants, fintechs, and developers who want to co-build experimental payment tech, from AI-driven checkout to agentic commerce, before it's a roadmap item anywhere else.
1999
Founded
50K+
merchants online with Autopay
300M
transactions a month
Top 500
World's Top Fintech Companies 2026
A regulated, PSD2-licensed European payment institution.
We don't sell a vision. We build things.
autopaylab isn't an R&D department locked in an ivory tower. It's a team that takes a real problem from a partner, sits down to write code, and in a few weeks has something that either works or doesn't. No pretending we have it all figured out upfront.
Why we're doing this at all
Because the most interesting things in payments won't get built at one table. We're opening up Autopay's infrastructure and engineers to anyone with an idea worth testing with us, instead of waiting for us to build it alone.
We research
We start from a real friction point in payments, not a whiteboard idea. Multiple agent protocols with no common integration layer was one of them.
We improve
We build on what already exists instead of starting from zero: existing checkout stacks, existing agent protocols like MCP, existing payment rails.
We build
What works ships as real code, not a slide deck. The Universal MCP Adapter is one adapter doing exactly this today.
We open it up
When something works, we open it to partners to build on, the way PayTalk's closed rollout works right now.
Three kinds of partner, one program
We're not building three separate tracks. Merchants, fintech and AI teams, and developers all sit at the same table and the same proof of concept can pull from all three.
E-commerce merchants
What you bring
- Production traffic and a real checkout to experiment on
What you get
- Early access to Autopay's AI-checkout and agentic-commerce builds
- Direct engineering support integrating them
- A launch-partner case study
Fintech and AI startups
What you bring
- A working product or model that needs a real payments rail
What you get
- Direct API access to Autopay's processing stack
- Co-development time with our engineers
- Where it fits, a warm intro to Autopay's merchant base for pilots
Developers and agencies
What you bring
- Build capacity: SDKs, integrations, agent tooling
What you get
- Early specs and sandbox access before public release
- A direct line to the team that owns the API
- Credit as a contributor when it ships
What we're building
We're not saying much yet. Here's what's queued up.
Universal MCP Adapter
Every agent framework speaks its own protocol, and every merchant runs a different backend. Wiring each combination together separately doesn't scale. The Universal MCP Adapter sits in the middle as a single translation layer: one side speaks whatever protocol an agent uses, the other speaks whatever a merchant's backend already exposes. Adding a new protocol or a new merchant means extending the adapter, not rebuilding the integration.
Universal MCP Adapter- One adapter instead of N×M integrations: agents and merchant backends each integrate once, against the adapter, not against each other.
- MCP is the first protocol it speaks; the header above shows the shape of that integration (a mocked create_payment tool call over MCP).
- Built to add protocols and merchant backends over time, without merchants or agent frameworks needing to change anything on their side.
Agentic Commerce
Reveal: October 2026
PayTalk
Timestamp: Q3 2026
mcp.autopaylab.com
Open beta since November 2025.
Apply to join the program
Tell us who you are and what you want to test. We read every application ourselves and reply directly, no automated screening.