Deel
Building AI tooling around MCP
MCP server and tool design connected AI-assisted internal products to useful context while keeping product and engineering judgment in the loop.
Problem
At Deel, internal teams needed AI-driven products and tools that could analyse data across customers and internal services. The goal was to give workers and teams useful context without exposing confidential implementation details.
Constraints
Global Payroll reporting involves complex financial and payroll data, large frontend datasets, and comparison workflows for around 1,000 customers. The interfaces must remain responsive, and the product work aims to reduce direct customer support.
Decision and implementation
I implemented an MCP server and tools used by internal teams, shaping their contracts and context around real analysis workflows. The work sits alongside AI-driven internal products, performance-sensitive frontend interfaces, and close collaboration with Product, Design, and customers.
Trade-off
AI tooling is only as dependable as its contracts and context. Performance work, production debugging, and incident investigation still require engineers to reason about the system instead of delegating judgment to a tool.
Impact and lesson
The MCP server and tools became internal developer products for analysing data across services. Code review and production investigation remain part of the workflow: I have identified regression causes even when AI-assisted debugging did not find them.