Model Context Protocol
Also called: MCP
By HubAI editorial team · Source checked: 24 October 2026What does it mean?
An open protocol for connecting AI applications to external capabilities such as tools, resources and prompts. It standardises parts of the connection, not the trustworthiness of everything connected.
Picture this
A compatible assistant connects to a document service through an MCP server. The service still needs authentication and must enforce which files the user can access.
Where you’ll come across it
Connecting assistants to business systems through reusable integrations. Check the server publisher, requested permissions and deployment model before connecting private material.
What it doesn’t mean
MCP support is not a security certification. A read connection and a tool allowed to send email carry different risks, even when both use the same protocol.
Draw the permission boundary
A short exercise · use fictional or non-sensitive material.
- Sketch a simple assistant that drafts a reply but cannot send it. Use invented recipients.
- Mark which tools it can read from, which it could write to and where approval is required.
- Try an instruction inside a sample document asking it to ignore those limits. Check that actual permissions still block the action.
Which safeguard is enforced by the application rather than a polite instruction?
Did that make it clearer?
Likes are a private on-page choice, not a public rating. For a correction or complaint, send the pre-filled form to our editorial inbox. Nothing is sent until you submit it.Check the source
Written and explained by HubAI. The reference below lets you go deeper; it is not an endorsement of a particular product.
Technical reference: modelcontextprotocol.io ↗