Skip to content
47% off: the complete course for $79 (reg. $149)Order now →
Agent Builder
Join
CurriculumPricingGlossaryFAQSign inAccount
Theme
Module 4Full course12 min read

Connecting MCP Servers to Your Agents

Install and configure MCP servers in popular agents, choose trustworthy servers, and troubleshoot connections.

Connecting your first MCP server feels like a superpower unlocking. One minute your agent can only see your files; the next it can read your GitHub issues, query your database, drive a browser, or search your team's docs. The configuration is usually a single command or a few lines of JSON.

The trouble is everything around those few lines. Which of the thousands of available servers are safe to run? Why does a server that works in one app fail silently in another? Where does the API key go so it doesn't end up in git? This lesson walks through connecting servers in Claude Code and other popular agents, both local and remote, then covers how to choose servers you can trust and a systematic way to troubleshoot connections when they break.

Finding servers

MCP servers come from a few places, roughly in descending order of trustworthiness:

  1. Official servers from the service provider. GitHub, Sentry, Notion, Linear, Stripe, and many others publish their own MCP servers, often as hosted remote servers you just sign in to. Start here whenever one exists.
  2. Reference servers from the MCP project, published under the @modelcontextprotocol npm scope, such as the filesystem and memory servers. They're well maintained but intended partly as examples.
  3. Your host's directory. Claude, ChatGPT, Cursor, and VS Code all offer curated lists of connectors or servers inside the app.
  4. The official MCP Registry and community lists. The MCP project runs a public registry of servers at registry.modelcontextprotocol.io, and there are large community-maintained lists. Useful for discovery, but listing doesn't mean vetting.

Before installing anything, recall the two transports from What Is MCP and Why It Matters: a local (stdio) server is a program the host launches on your machine; a remote (Streamable HTTP) server lives at a URL. Configuration differs accordingly. Local servers need a command to run; remote servers need a URL and usually authentication.