Skills and MCP

A skill packages reusable instructions, references, and sometimes scripts for a repeated kind of work. It teaches an agent a process without retraining the model.

For example, a pull-request review skill can define the review checklist, severity levels, output format, and checks to run. The same process can be reused across many repositories.

MCP connects capabilities

MCP, the Model Context Protocol, gives AI applications a standard way to connect to external tools and data.

The main participants are:

• the host, which is the AI application
• an MCP client managed by the host
• an MCP server, which exposes capabilities

An MCP server can expose tools that perform actions, resources that provide data, and prompts that provide reusable interaction templates.

A simplified flow looks like this:

1. The host connects to the server.
2. The client discovers the capabilities the server exposes.
3. The model selects a relevant capability.
4. The host applies permissions and may request approval.
5. The server performs the operation and returns a result.

Process and capability are different

A skill explains how to review a pull request. A GitHub MCP server can supply the pull request, files, and action for posting a comment.

Either can exist without the other. You can follow a review skill using local Git commands. You can use a GitHub MCP tool without having a good review process. Together they provide know-how and access.

Standard does not mean trusted

MCP standardizes communication. It does not make a server safe, correct, or sandboxed. A server may expose broad tools, mishandle credentials, return malicious content, or perform a different action than its name suggests.

Review the server source or provider, exposed capabilities, authentication scopes, data flow, and approval behavior before connecting it to important systems.

Skills deserve review too. A skill can include scripts or tell an agent to run commands. Treat installing one like adding code or automation to your development environment.

Continue with the MCP Course to understand clients, servers, tools, resources, prompts, configuration, and trust boundaries.