Technology

Tools & MCP

The connections that let AI do things — read a calendar, query a database, send a draft — with MCP emerging as a standard plug for those connections.

By itself, a language model can only produce text. Tools change that: they are the connectors through which the model can search the web, read a CRM record, check a calendar or file a document. MCP — the Model Context Protocol — is an emerging open standard for these connections, aiming to do for AI integrations what USB did for peripherals: one standard plug instead of bespoke wiring for every tool.

Executives should care because tools are where capability meets risk. Every tool you grant an agent is a power it wields: read access to a system, or worse, write access. The governance question 'which tools may this agent use?' is really 'what is this agent allowed to do?' — and should be answered with the same care as system access for employees.

Concrete example: a meeting-prep agent with calendar-read, CRM-read and document-search tools can produce excellent briefs with zero ability to harm. Give it email-send, and the risk profile changes completely.

Standardisation matters commercially too: MCP-style standards reduce vendor lock-in and make integrations cheaper to build and swap.