Connecting Tools With MCP
Out of the box an agent can read files and run shell commands. MCP (the Model Context Protocol) is the open standard that lets it reach further: into your database, your GitHub issues, your browser, your design files. An MCP server exposes a set of tools, and once connected, your agent can call them as naturally as it edits a file.
Step 1: Understand what a server gives you
Each MCP server adds capabilities. A GitHub server lets the agent read and comment on issues and PRs. A Playwright server lets it drive a real browser to test your app. A Postgres server lets it query your database safely. You pick the ones that match your stack.
| MCP server | What the agent can do |
|---|---|
| GitHub | Read issues, open and review PRs |
| Playwright | Click through your app in a real browser |
| Postgres | Run read-only queries against your DB |
| Filesystem | Reach files outside the project root, scoped |
Step 2: Add a server
In Claude Code you add a server with the mcp command. In Cursor you add it in settings under MCP. Either way you point at a server package or URL and the agent picks up the new tools.
Step 3: Use the new powers in a prompt
Once connected, you just ask. The agent figures out which tool to call. For example, with the GitHub and Playwright servers attached, you can hand it an issue and let it close the loop end to end.
Read GitHub issue #214, reproduce the bug by clicking through the app
with the browser tool, fix it, and confirm the steps in the issue now pass.