1 min lesson
Review an MCP server
Decide whether to approve a server by checking its operator, permissions, data and configuration.
Step 1 of 2
Review each MCP serverownership, access, data and approval
- Require admin approval for every server the company permits.
- Prefer read-only access when the agent only needs context. Grant write access only for a specific action the workflow requires.
- Assess the server's data handling separately from Cursor's model data path.
- Log MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. calls so the team can see which systems an agent used and when.
- Who hosts it?
- Identify whether your team, a vendor or a community maintainer operates the server.
- Read or write?
- Start with read access and require a reason for every write action.
- What data flows through?
- List any secrets, personal information or regulated data the server may expose.
- On the allowlist?
- Block the server until an admin has approved its configuration.
Review the operator, permissions, data and approval before connecting a server.
A team proposes a community-hosted issue tracker server with write access. The workflow only needs to read ticket details, which may include customer information. No admin has reviewed the server.
Block the connection for now. The team can reconsider it after an admin reviews the operator and data handling, and the requested access is reduced to read-only.
Learn more
Optional practice
Choose the right MCP access
QA vendor-hosted Confluence MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. server will let agents read runbooks. It requests read and write access, and no admin has reviewed it. What should the team do?