AI integration & MCP

Connect the model.Control the action.

Connect AI to the systems that run your organisation. Define the interfaces, validate the inputs and enforce permission where an action executes.

Explore the capability
ENGINEERING FOCUS
01Validated request
02Authorised tool
03Checked result

The interesting part is the boundary.

Reading a record, preparing a change and committing it are different operations. A model can request a tool call; the application decides whether it is authorised.

01

Design the interface.

Use APIs, Model Context Protocol tools or event-driven integration where appropriate. Define inputs, outputs and failure behaviour.

02

Scope the access.

Authenticate requests, minimise privileges and keep credentials out of prompts. Treat tool output as data, not new authority.

03

Make changes traceable.

Record the request, approval and resulting action. Handle timeouts, duplicates and partial completion explicitly.

A possible workflowIllustrative example.

Read a customer record, draft an update and require separate authorisation before writing it back.

A protocol is not a security guarantee. Implement and test consent, authorisation, credential handling and network restrictions.

A closer look

Good questions.
Straight answers.

What does MCP add?

A standard way for models and tools to exchange capabilities and requests. It does not remove the need for application security.

What about software without an API?

Controlled computer-use workflows may be an option after assessing the interface and failure modes.

Technical reference: MCP: security best practices (opens in a new tab)