Skip to lesson
Exit
MCP Fundamentals1 / 2

2 min lesson

Tools & MCPs lists every connected server

Connect "Start with this check: The server exposes the expected tool names" to the decision you would make.

Step 1 of 2

The case: you can name when an 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 earns its place, and the trust-boundary shift that approving one forces. Start with this check: The server exposes the expected tool names. Walk it once against a real situation so you can defend the call, not just assert it.

Name the MCP servers before you approve another

0:20 · narrated
Read this demo as text
  1. Tools and MCPs is where connected servers live. Mobbin, PostHog, Railway — each row is a real capability boundary. Name the tool you need before you approve another server.
Tools & MCPs lists every connected server — name the capability before you approve another one.

Practice next: Open Tools & MCPs in your project and list which servers are live before you add another.

Simulated Cursor 3.12 (macOS, light) — beta educational reconstruction, not the real product.

Learn more

Full explanation

Model Context Protocol server

One MCP server — does a Linear connection earn a place in the agent?
Model Context Protocol server
SayThe agent keeps asking me to paste the acceptance criteria. They already live on the Linear issue, and I want it to read the issue itself.
DoName the capability first: read one Linear issue and its comments. That, not 'connect Linear', is the job a Model Context Protocol (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 would do here.
DoSort what the server exposes into three primitives. A tool is a function the agent calls, like search issues or read issue 482. A resource is structured data it can reference. A prompt is a templated workflow the server ships. I need one read tool, not the create-issue and update-issue tools sitting beside it.
SeeNaming the specific tool, not 'a Linear connection', tells you exactly what the agent can invoke once you approve it.
DoOpen mcp.json and check two things. The transport: a local stdio process Cursor runs, or a remote server reached over HTTP. And the credential: an env var or header pulled in with a placeholder like ${env:LINEAR_TOKEN}, never a token pasted inline.
SeeA remote server holding a broad API token reaches further than a local stdio process scoped to one workspace. Same feature, wider blast radiusHow much breaks if a change goes wrong; the scope of potential damage. Press Enter for the full definition..
DoApprove the read and search tools, leave create and update unapproved, then run the task.
SeeThe trust boundary just moved. The agent can pull real issue text now, and whatever that issue says becomes context that can steer the edits it proposes. That shift is what approving any 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 forces, so you grant the narrowest capability that does the job.
Learn more

Optional practice

Practice: Tools & MCPs lists every connected server

QYou've applied it. What actually proves the work is done, not just a plausible answer?