Adfin | Adfin releases beta MCP server for AI clients

We believe finance is riddled with inefficient, manual processes. We combine payments infrastructure innovation with agentic AI to transform the financial operations of SMBs.

We always want to always make sure we’re on the leading edge of technology. Recently we hosted a hackathon to tackle a fundamental challenge in AI-driven fintech. Our goal? To develop an interface that allows you to use Adfin, from anywhere, anytime - via an LLM (large language model eg. Claude).

‍

For anyone unfamiliar with the concept of MCP (Model Context Protocol), here’s a quick introduction:

MCP is an open protocol that standardizes how applications provide context to LLMs. Think of MCP like a USB-C port for AI applications. Just as USB-C provides a standardized way to connect your devices to various peripherals and accessories, MCP provides a standardized way to connect AI models to different data sources and tools. (https://modelcontextprotocol.io/introduction)

‍

What can you do with an MCP server?

Our idea was, if we can build an MCP Server - an Adfin customer can essentially talk to Adfin, from any LLM that supports MCP (eg. Claude desktop) - and have a new, decoupled, interface for managing all of their cashflow.

What we've built

Architecture for Adfin's MCP Server

We’ve released an MCP server that allows use of Claude desktop for almost all of the operations in the Adfin platform.

‍Here’s some examples:

🚀 Ask Claude about your Credit control, such as outstanding invoices (new APIs!).

🚀 Ask Claude to create you new customers in Adfin.

🚀 Ask Claude to create an invoice via text, send it to Adfin and activate reminders for it.

🚀 Upload an invoice to Claude, send it to Adfin and activate it.

🚀 Ask Claude to check the state of your invoices.

Create an invoice for individuals, straight from your LLM.

‍

Claude can generate a quick summary of your credit risk profile

‍

You can check out the MCP server here: https://github.com/Adfin-Engineering/mcp-server-adfin.

We have also published a video demonstration for anyone who just wants to see how it works, without the technical setup.

Interested in giving this a go, but don't know where to start? Shortly we will release some content around how to get this going on your own machine.

‍

Challenges

It wouldn’t be fair to highlight some of the challenges we’ve had with getting started with the MCP, so we’ve included some details on this. While most of this is down to the maturity of the MCP protocol, it shows that we still have some way to go before MCP can be mainstream.

‍

❌ Challenge 1: Lack of native oAuth:

Despite all the fun tools, the big limiting factor across for us, was the lack of native interface MCP for oAuth (which Adfin’s API relies upon, by design). Adfin’s APIs are predominantly driven by user level access, driven by user bearer tokens. There doesn’t seem to be a repeatable way to implement this today.

❌ Challenge 2: Protocol limitations

While we were able to successfully connect to the Adfin MCP server from the console, the value of such a server comes from an LLM having access to the tools to use as they see fit. However, the current limitations of the MCP protocol around authentication don't allow this, and also the output presented by the LLM to the user seems to be aimed at people with a technical background.

❌ Challenge 3: Spring framework limited our bridge between stdio and http.

A bridge was required for interoperability with Claude Sonnet Desktop app, which only supports stdio (for now). Unfortunately we were unable to make any of them work, it is not clear what is the issue. Could be either the bridge, either something on the spring boot MCP server, but we weren’t able to pinpoint the cause exactly. Eventually ditched the bridge as we changed strategy on how we integrate with Adfin APIs.

‍

Our conclusion

Despite encountering challenges, our journey into integrating the MCP protocol with Adfin's powerful payment infrastructure and AI capabilities has been incredibly exciting. The current limitations—particularly around authentication and infrastructure compatibility—highlight the early-stage nature of MCP, yet also underscore the immense potential this technology holds.

We're committed to continuing our exploration and innovation in this space. As MCP matures and evolves, we eagerly anticipate addressing these initial hurdles and unlocking even greater possibilities. Our experience thus far reinforces our belief in the transformative potential of standardized, context-aware interfaces in fintech.

We look forward to actively contributing to, and participating in, the future developments of MCP, ensuring that Adfin remains at the forefront of AI-driven financial innovation.

TL;DR: It’s very cool. Like really really cool. But it’s not quite as quick to get going as X/Twitter/LinkedIn would have you believe.

Adfin releases beta MCP server for AI clients

Read more

Meet Chris Carty our new Solutions & Delivery Specialist

Meet Toby Piper, our new Software Engineer

How Karolina made getting paid feel effortless