Get a quote

MCP security checklist: 12 checks before you connect AI to company data

Published 2026-10-14. 6 min read.

An MCP server is a new way into your systems. These are the checks we run on every connector we build or audit.

Credentials and access

  1. Each server uses its own credential, never a personal admin token.
  2. Scopes are the minimum the tools need.
  3. Secrets live in a secret store or environment, not in shared config files.
  4. If users have different permissions, the server acts as each user through OAuth.

Tools

  1. Read is the default. Every write tool is a deliberate decision.
  2. Destructive or financial actions require confirmation or are not exposed.
  3. Inputs are validated on the server, not trusted from the model.
  4. Result size and query time are capped on the server.

Prompt injection

  1. Tools that read untrusted content (emails, tickets, web pages) are not combined with powerful write tools in the same session without review.
  2. Tool outputs are treated as data. The server never executes instructions found in content.

Operations

  1. Every tool call is logged with user, tool, inputs and result status.
  2. Someone owns the server: updates, alerts and a way to switch it off.

Want us to run this checklist on your setup? See MCP setup and audit.

Related guides

Tell us which system your team wants AI to work with.

Fill in a 5-minute brief. Within 48 hours you get a tool list, a fixed price and an honest answer on whether you need us at all. No meetings needed.

Get a fixed quote