Company
Building the AI-Powered Future of Payment Intelligence: Insights from Our Agent Hackathon
October 9, 2025
October 9, 2025



At the tail end of Q3, the globally-distributed Pagos flock met up for an in-person offsite. For our engineering team, these meetups provide a rare opportunity for working, planning, and experimenting together. At our June offsite, we used this time to explore AI-assisted development through a Code Retreat using Cursor. This time, we took the next evolutionary step and hosted an Agent Hackathon, exploring how AI agents powered by the latest frameworks can autonomously interact with payment data to solve complex business problems.
The Tools: AI Frameworks for Agent Development
To drive engineering productivity and deliver value at scale, we've been leveraging AI tools like Cursor for a while now. This hackathon gave us the chance to go deeper—exploring frameworks designed specifically for building autonomous agents that plan, analyze, and act with minimal human input:
Vercel AI SDK 5
Released in July, Vercel’s latest AI SDK offers features designed for building maintainable agent UIs, which align perfectly with our goals. A few features stood out:
Dynamic support for rendering tools from MCP servers whose definitions aren't known at build time
Streaming tool execution so users could see real-time updates as agents worked through complex payment analyses
Framework flexibility, allowing teams to build with React, Vue, Svelte, or Angular
Ultimately, AI SDK 5 let our teams focus on building innovative payment solutions rather than wrestling with UI implementation details.
Strands Agents SDK
Strands Agents SDK by AWS is an open-source, model-driven framework for building AI agents. Rather than requiring complex orchestration logic, Strands lets agents leverage the reasoning capabilities of modern LLMs to autonomously plan and execute tasks.
What made Strands appealing for our hackathon goals:
Minimal code required - We could create powerful agents with just a prompt and a list of tools
Real-world usage - It's being used in production by services like Amazon Q Developer and AWS Glue
Model flexibility - Support for Amazon Bedrock, Anthropic, OpenAI, and more via LiteLLM gave our teams options
Native MCP support - Native support for Model Context Protocol (MCP) servers made it seamless to integrate with our hosted MCP server preview
For us, it lowered the barrier to entry and let our teams prototype fast.
MCP at Pagos: From BIN Data to Broader Platform Access
Since announcing our first MCP server in June, we've continued expanding what's possible with AI-powered payment intelligence. While our open source MCP implementation already enables developers to query BIN data through their AI assistants, we're working on a hosted version that aims to expand capabilities significantly, for example:
Query transaction data across multiple dimensions
Analyze payment trends and anomalies in real-time
Access more of our payment optimization toolkit
Tap into cross-processor analytics and benchmarking data
Generate insights from complex payment patterns
For our hackathon, teams got to experiment with a preview of our hosted MCP server, using the same standardized protocol but with access to more data.
Winning Projects: Where Creativity Meets Payment Intelligence
The Make it Fun Award: Payments Tarot Cards 🔮
Building on our tradition of encouraging creative AI experiments (remember the emoji Game of Life from our last retreat?), we awarded a prize for the most imaginative use of payment data. The winning team created PaymentMystic, an AI fortune teller that uses our preview hosted MCP server to gather a customer's transaction data from the previous week and transform it into personalized Tarot cards. Each card represents different aspects of their payment health:
The Star = A spike in transaction approval rate
The Fool = A change in Chargeback Rate
The Magician = A spike in total transaction count
The LLM analyzes payment patterns to create surprisingly insightful (and entertaining) fortunes like "The Three of Cups reversed suggests your subscription services are draining your treasury—perhaps it's time to audit those monthly charges."

Best Overall: Autonomous Payment Analysis with Sandboxed Execution 📊
Our chatbot, Pagos AI, uses OpenAI's Assistants API to write Python code for complex payment data analysis. With that API being deprecated, we need to explore alternatives that maintain this powerful capability while adding the benefits of modern agent frameworks.
The winning team built DataWeaver, a solution that combines all of the following features:
Vercel's AI SDK for streaming responses and tool management
Sandboxed Javascript execution for secure code generation and execution
Our hosted MCP server preview for comprehensive real-time data access
Strands Agents for orchestrating complex multi-step analyses
What made this project stand out:
Security first: All generated code runs in isolated sandboxes
Streaming insights: Users see results as they're computed, not after
Visual outputs: Automatic generation of charts and visualizations
Reproducible analyses: All code is saved and can be re-run or modified
Example workflow demonstrated:
User: "Show me how our approval rates correlate with transaction amounts for different card brands over the last quarter"
Agent:
1. Queries comprehensive transaction data via hosted MCP server
2. Generates Python code for correlation analysis
3. Creates interactive visualizations
4. Streams insights as they're discovered
5. Provides actionable recommendations
The Future of AI-Powered Payment Intelligence
This hackathon gave us the opportunity to explore how AI agents can transform payment data analysis. We experimented with ways to move beyond using AI as a development assistant to building systems where AI agents could potentially analyze payments data autonomously and provide actionable insights. The combination of AI frameworks like these, standardized protocols like MCP, and deep payment intelligence opens up exciting possibilities we're eager to continue exploring.
From Open Source to Hosted: Expanding Our MCP Capabilities
When we launched our MCP server, we committed to expanding our MCP implementation in the coming months. This hackathon gave us the opportunity to preview where we're heading.
Our planned hosted MCP server builds on the foundation of our open source implementation, with the goal of significantly expanding what's possible. This includes broader access to more of our payment intelligence platform, such as live transaction data and cross-processor benchmarking, and the ability to handle complex queries across large datasets. We also aim to have a simplified setup, allowing non-technical users who don’t necessarily want to run our MCP server locally to access their payments data within their preferred AI assistant.
The vision is for AI assistants to be able to answer questions like:
"Which processors are seeing increased decline rates for subscription renewals this week?"
"What's the correlation between transaction size and approval rates across all our payment methods?"
"Show me anomalous patterns in our authorization data that might indicate fraud"
"Compare our performance metrics against industry benchmarks for similar merchants"
We're excited about the potential to provide comprehensive payment insights through natural conversation with AI.
Why This Matters
The convergence of AI agents and payment intelligence opens up exciting new possibilities for how businesses could optimize payment. We're exploring futures that could include:
Optimization agents that continuously monitor and suggest routing improvements
Predictive analytics that could help identify issues before they impact revenue
Natural language interfaces that make payment insights accessible to anyone in an organization
Cross-platform intelligence that connects payment data with other business systems
These are the kinds of capabilities we're working towards, and our hackathon gave us valuable insights into what's possible.
Join Us!
From our AI-assisted Code Retreat to this Agent Hackathon exploring autonomous systems, we’re just getting started. Follow along as we continue exploring what’s possible when AI meets payment intelligence!
Whether you're a developer looking to build the next generation of payment tools or a business seeking to unlock the value in your payment data, we're here to help you succeed. Check out our open source MCP server, or reach out today to request early beta access to our hosted MCP server.
At the tail end of Q3, the globally-distributed Pagos flock met up for an in-person offsite. For our engineering team, these meetups provide a rare opportunity for working, planning, and experimenting together. At our June offsite, we used this time to explore AI-assisted development through a Code Retreat using Cursor. This time, we took the next evolutionary step and hosted an Agent Hackathon, exploring how AI agents powered by the latest frameworks can autonomously interact with payment data to solve complex business problems.
The Tools: AI Frameworks for Agent Development
To drive engineering productivity and deliver value at scale, we've been leveraging AI tools like Cursor for a while now. This hackathon gave us the chance to go deeper—exploring frameworks designed specifically for building autonomous agents that plan, analyze, and act with minimal human input:
Vercel AI SDK 5
Released in July, Vercel’s latest AI SDK offers features designed for building maintainable agent UIs, which align perfectly with our goals. A few features stood out:
Dynamic support for rendering tools from MCP servers whose definitions aren't known at build time
Streaming tool execution so users could see real-time updates as agents worked through complex payment analyses
Framework flexibility, allowing teams to build with React, Vue, Svelte, or Angular
Ultimately, AI SDK 5 let our teams focus on building innovative payment solutions rather than wrestling with UI implementation details.
Strands Agents SDK
Strands Agents SDK by AWS is an open-source, model-driven framework for building AI agents. Rather than requiring complex orchestration logic, Strands lets agents leverage the reasoning capabilities of modern LLMs to autonomously plan and execute tasks.
What made Strands appealing for our hackathon goals:
Minimal code required - We could create powerful agents with just a prompt and a list of tools
Real-world usage - It's being used in production by services like Amazon Q Developer and AWS Glue
Model flexibility - Support for Amazon Bedrock, Anthropic, OpenAI, and more via LiteLLM gave our teams options
Native MCP support - Native support for Model Context Protocol (MCP) servers made it seamless to integrate with our hosted MCP server preview
For us, it lowered the barrier to entry and let our teams prototype fast.
MCP at Pagos: From BIN Data to Broader Platform Access
Since announcing our first MCP server in June, we've continued expanding what's possible with AI-powered payment intelligence. While our open source MCP implementation already enables developers to query BIN data through their AI assistants, we're working on a hosted version that aims to expand capabilities significantly, for example:
Query transaction data across multiple dimensions
Analyze payment trends and anomalies in real-time
Access more of our payment optimization toolkit
Tap into cross-processor analytics and benchmarking data
Generate insights from complex payment patterns
For our hackathon, teams got to experiment with a preview of our hosted MCP server, using the same standardized protocol but with access to more data.
Winning Projects: Where Creativity Meets Payment Intelligence
The Make it Fun Award: Payments Tarot Cards 🔮
Building on our tradition of encouraging creative AI experiments (remember the emoji Game of Life from our last retreat?), we awarded a prize for the most imaginative use of payment data. The winning team created PaymentMystic, an AI fortune teller that uses our preview hosted MCP server to gather a customer's transaction data from the previous week and transform it into personalized Tarot cards. Each card represents different aspects of their payment health:
The Star = A spike in transaction approval rate
The Fool = A change in Chargeback Rate
The Magician = A spike in total transaction count
The LLM analyzes payment patterns to create surprisingly insightful (and entertaining) fortunes like "The Three of Cups reversed suggests your subscription services are draining your treasury—perhaps it's time to audit those monthly charges."

Best Overall: Autonomous Payment Analysis with Sandboxed Execution 📊
Our chatbot, Pagos AI, uses OpenAI's Assistants API to write Python code for complex payment data analysis. With that API being deprecated, we need to explore alternatives that maintain this powerful capability while adding the benefits of modern agent frameworks.
The winning team built DataWeaver, a solution that combines all of the following features:
Vercel's AI SDK for streaming responses and tool management
Sandboxed Javascript execution for secure code generation and execution
Our hosted MCP server preview for comprehensive real-time data access
Strands Agents for orchestrating complex multi-step analyses
What made this project stand out:
Security first: All generated code runs in isolated sandboxes
Streaming insights: Users see results as they're computed, not after
Visual outputs: Automatic generation of charts and visualizations
Reproducible analyses: All code is saved and can be re-run or modified
Example workflow demonstrated:
User: "Show me how our approval rates correlate with transaction amounts for different card brands over the last quarter"
Agent:
1. Queries comprehensive transaction data via hosted MCP server
2. Generates Python code for correlation analysis
3. Creates interactive visualizations
4. Streams insights as they're discovered
5. Provides actionable recommendations
The Future of AI-Powered Payment Intelligence
This hackathon gave us the opportunity to explore how AI agents can transform payment data analysis. We experimented with ways to move beyond using AI as a development assistant to building systems where AI agents could potentially analyze payments data autonomously and provide actionable insights. The combination of AI frameworks like these, standardized protocols like MCP, and deep payment intelligence opens up exciting possibilities we're eager to continue exploring.
From Open Source to Hosted: Expanding Our MCP Capabilities
When we launched our MCP server, we committed to expanding our MCP implementation in the coming months. This hackathon gave us the opportunity to preview where we're heading.
Our planned hosted MCP server builds on the foundation of our open source implementation, with the goal of significantly expanding what's possible. This includes broader access to more of our payment intelligence platform, such as live transaction data and cross-processor benchmarking, and the ability to handle complex queries across large datasets. We also aim to have a simplified setup, allowing non-technical users who don’t necessarily want to run our MCP server locally to access their payments data within their preferred AI assistant.
The vision is for AI assistants to be able to answer questions like:
"Which processors are seeing increased decline rates for subscription renewals this week?"
"What's the correlation between transaction size and approval rates across all our payment methods?"
"Show me anomalous patterns in our authorization data that might indicate fraud"
"Compare our performance metrics against industry benchmarks for similar merchants"
We're excited about the potential to provide comprehensive payment insights through natural conversation with AI.
Why This Matters
The convergence of AI agents and payment intelligence opens up exciting new possibilities for how businesses could optimize payment. We're exploring futures that could include:
Optimization agents that continuously monitor and suggest routing improvements
Predictive analytics that could help identify issues before they impact revenue
Natural language interfaces that make payment insights accessible to anyone in an organization
Cross-platform intelligence that connects payment data with other business systems
These are the kinds of capabilities we're working towards, and our hackathon gave us valuable insights into what's possible.
Join Us!
From our AI-assisted Code Retreat to this Agent Hackathon exploring autonomous systems, we’re just getting started. Follow along as we continue exploring what’s possible when AI meets payment intelligence!
Whether you're a developer looking to build the next generation of payment tools or a business seeking to unlock the value in your payment data, we're here to help you succeed. Check out our open source MCP server, or reach out today to request early beta access to our hosted MCP server.
At the tail end of Q3, the globally-distributed Pagos flock met up for an in-person offsite. For our engineering team, these meetups provide a rare opportunity for working, planning, and experimenting together. At our June offsite, we used this time to explore AI-assisted development through a Code Retreat using Cursor. This time, we took the next evolutionary step and hosted an Agent Hackathon, exploring how AI agents powered by the latest frameworks can autonomously interact with payment data to solve complex business problems.
The Tools: AI Frameworks for Agent Development
To drive engineering productivity and deliver value at scale, we've been leveraging AI tools like Cursor for a while now. This hackathon gave us the chance to go deeper—exploring frameworks designed specifically for building autonomous agents that plan, analyze, and act with minimal human input:
Vercel AI SDK 5
Released in July, Vercel’s latest AI SDK offers features designed for building maintainable agent UIs, which align perfectly with our goals. A few features stood out:
Dynamic support for rendering tools from MCP servers whose definitions aren't known at build time
Streaming tool execution so users could see real-time updates as agents worked through complex payment analyses
Framework flexibility, allowing teams to build with React, Vue, Svelte, or Angular
Ultimately, AI SDK 5 let our teams focus on building innovative payment solutions rather than wrestling with UI implementation details.
Strands Agents SDK
Strands Agents SDK by AWS is an open-source, model-driven framework for building AI agents. Rather than requiring complex orchestration logic, Strands lets agents leverage the reasoning capabilities of modern LLMs to autonomously plan and execute tasks.
What made Strands appealing for our hackathon goals:
Minimal code required - We could create powerful agents with just a prompt and a list of tools
Real-world usage - It's being used in production by services like Amazon Q Developer and AWS Glue
Model flexibility - Support for Amazon Bedrock, Anthropic, OpenAI, and more via LiteLLM gave our teams options
Native MCP support - Native support for Model Context Protocol (MCP) servers made it seamless to integrate with our hosted MCP server preview
For us, it lowered the barrier to entry and let our teams prototype fast.
MCP at Pagos: From BIN Data to Broader Platform Access
Since announcing our first MCP server in June, we've continued expanding what's possible with AI-powered payment intelligence. While our open source MCP implementation already enables developers to query BIN data through their AI assistants, we're working on a hosted version that aims to expand capabilities significantly, for example:
Query transaction data across multiple dimensions
Analyze payment trends and anomalies in real-time
Access more of our payment optimization toolkit
Tap into cross-processor analytics and benchmarking data
Generate insights from complex payment patterns
For our hackathon, teams got to experiment with a preview of our hosted MCP server, using the same standardized protocol but with access to more data.
Winning Projects: Where Creativity Meets Payment Intelligence
The Make it Fun Award: Payments Tarot Cards 🔮
Building on our tradition of encouraging creative AI experiments (remember the emoji Game of Life from our last retreat?), we awarded a prize for the most imaginative use of payment data. The winning team created PaymentMystic, an AI fortune teller that uses our preview hosted MCP server to gather a customer's transaction data from the previous week and transform it into personalized Tarot cards. Each card represents different aspects of their payment health:
The Star = A spike in transaction approval rate
The Fool = A change in Chargeback Rate
The Magician = A spike in total transaction count
The LLM analyzes payment patterns to create surprisingly insightful (and entertaining) fortunes like "The Three of Cups reversed suggests your subscription services are draining your treasury—perhaps it's time to audit those monthly charges."

Best Overall: Autonomous Payment Analysis with Sandboxed Execution 📊
Our chatbot, Pagos AI, uses OpenAI's Assistants API to write Python code for complex payment data analysis. With that API being deprecated, we need to explore alternatives that maintain this powerful capability while adding the benefits of modern agent frameworks.
The winning team built DataWeaver, a solution that combines all of the following features:
Vercel's AI SDK for streaming responses and tool management
Sandboxed Javascript execution for secure code generation and execution
Our hosted MCP server preview for comprehensive real-time data access
Strands Agents for orchestrating complex multi-step analyses
What made this project stand out:
Security first: All generated code runs in isolated sandboxes
Streaming insights: Users see results as they're computed, not after
Visual outputs: Automatic generation of charts and visualizations
Reproducible analyses: All code is saved and can be re-run or modified
Example workflow demonstrated:
User: "Show me how our approval rates correlate with transaction amounts for different card brands over the last quarter"
Agent:
1. Queries comprehensive transaction data via hosted MCP server
2. Generates Python code for correlation analysis
3. Creates interactive visualizations
4. Streams insights as they're discovered
5. Provides actionable recommendations
The Future of AI-Powered Payment Intelligence
This hackathon gave us the opportunity to explore how AI agents can transform payment data analysis. We experimented with ways to move beyond using AI as a development assistant to building systems where AI agents could potentially analyze payments data autonomously and provide actionable insights. The combination of AI frameworks like these, standardized protocols like MCP, and deep payment intelligence opens up exciting possibilities we're eager to continue exploring.
From Open Source to Hosted: Expanding Our MCP Capabilities
When we launched our MCP server, we committed to expanding our MCP implementation in the coming months. This hackathon gave us the opportunity to preview where we're heading.
Our planned hosted MCP server builds on the foundation of our open source implementation, with the goal of significantly expanding what's possible. This includes broader access to more of our payment intelligence platform, such as live transaction data and cross-processor benchmarking, and the ability to handle complex queries across large datasets. We also aim to have a simplified setup, allowing non-technical users who don’t necessarily want to run our MCP server locally to access their payments data within their preferred AI assistant.
The vision is for AI assistants to be able to answer questions like:
"Which processors are seeing increased decline rates for subscription renewals this week?"
"What's the correlation between transaction size and approval rates across all our payment methods?"
"Show me anomalous patterns in our authorization data that might indicate fraud"
"Compare our performance metrics against industry benchmarks for similar merchants"
We're excited about the potential to provide comprehensive payment insights through natural conversation with AI.
Why This Matters
The convergence of AI agents and payment intelligence opens up exciting new possibilities for how businesses could optimize payment. We're exploring futures that could include:
Optimization agents that continuously monitor and suggest routing improvements
Predictive analytics that could help identify issues before they impact revenue
Natural language interfaces that make payment insights accessible to anyone in an organization
Cross-platform intelligence that connects payment data with other business systems
These are the kinds of capabilities we're working towards, and our hackathon gave us valuable insights into what's possible.
Join Us!
From our AI-assisted Code Retreat to this Agent Hackathon exploring autonomous systems, we’re just getting started. Follow along as we continue exploring what’s possible when AI meets payment intelligence!
Whether you're a developer looking to build the next generation of payment tools or a business seeking to unlock the value in your payment data, we're here to help you succeed. Check out our open source MCP server, or reach out today to request early beta access to our hosted MCP server.
Share this Blog Post
Share this Blog Post
Latest Blog Posts

Building the AI-Powered Future of Payment Intelligence: Insights from Our Agent Hackathon
Building the AI-Powered Future of Payment Intelligence: Insights from Our Agent Hackathon
Building the AI-Powered Future of Payment Intelligence: Insights from Our Agent Hackathon

Defining Payments Anomalies for Yourself
Defining Payments Anomalies for Yourself
Defining Payments Anomalies for Yourself

Building a Resilient Subscription Strategy with Pagos
Building a Resilient Subscription Strategy with Pagos
Building a Resilient Subscription Strategy with Pagos
Subscribe to our Blog
Subscribe to
our Blog
Subscribe to our Blog
By submitting, you are providing your consent for future communication in accordance with the Pagos Privacy Policy.