Skip to main content
Vibe Code Forum

Article: MCP explained: letting your AI tool use other tools

MCP (Model Context Protocol) is an open standard, introduced by Anthropic in late 2024 and since adopted widely, that lets AI tools connect to other tools and data sources in a common way. Think of it as a universal plug: a service publishes an MCP server, and any AI tool that supports MCP (many editors, terminal agents and desktop chat apps do) can use it.

What it makes possible

Without MCP, your AI only knows what you paste in. With MCP servers connected, it can, for example:

  • read up-to-date documentation for a library instead of guessing from old training data,
  • look at your database's tables or your hosting logs,
  • open a real browser, click through your app and take screenshots,
  • read and create issues in GitHub or tasks in your project tracker.

How it works, briefly

  • An MCP server is a small program that offers tools (actions such as "run query") and resources (data such as files or records).
  • Your AI app (the client) connects to it, either by starting it on your computer or over the internet.
  • The AI decides when to use a tool, and most clients ask your permission first.

Adding one

Each AI tool has its own settings screen or config file for MCP servers. The server's documentation shows the snippet to add, usually a command to run plus any keys it needs. Restart the tool, then ask: "Which MCP tools do you have available?"

Use them carefully

  • Only install servers from sources you trust. An MCP server runs with your permissions and can read what you give it access to.
  • Give the least access that works. Read-only database access for exploring, a test project rather than production.
  • Watch for prompt injection. Content the AI reads through a tool (a web page, an issue, an email) can contain hidden instructions. Keep approval prompts on for actions that change or delete things.
  • Keys stay in config, not in chat. Put tokens in the server's settings or environment, never in a prompt.

Start with one or two servers that solve a real annoyance, such as documentation lookup or browser testing, rather than connecting everything at once.

Read next: Chat vs agent: what changes when the AI can run commands and APIs and API keys: how apps talk to each other.

More in Tools explained