Challenge
Even when a company wants to embed AI into an existing service, the business logic and data are locked inside the existing system and cannot be handled safely by AI as-is. Rebuilding is not realistic, so a path was needed to advance AI adoption while keeping existing assets in place.
Approach
- 01 Took inventory of the existing service's functions and data, and defined the scope and permission boundaries AI would operate within
- 02 Implemented an MCP server per function, exposing existing APIs and databases in a form AI can call safely
- 03 Designed the permitted data and operations under least privilege, on the premise of access control and audit logs
- 04 Kept AI output under human review, with senior engineers standing behind critical decisions
- 05 Left the existing system untouched, expanding AI use incrementally by adding the connection layer
Key success factors
- Leveraging existing assets
- Rather than rebuilding the system, we added an MCP connection layer, advancing AI adoption while preserving the existing business logic.
- Guaranteeing permissions and auditability
- By designing the scope AI can handle under least privilege and making every operation auditable, we embedded AI safely.
Solution
We made the existing service's functions and data accessible to AI via the Model Context Protocol (MCP), building a connection layer that guarantees permissions and auditability. Without rebuilding the existing system, we created a state in which AI agents can safely use internal tools and data.
Facing a similar challenge?
Whatever your industry or scale, let's start with a conversation.
Contact form
