Claude

MCP in Claude Code

MCP connects Claude Code to external tools and data sources like Linear, GitHub, and documentation servers. This video covers adding servers, scoping them for your team, and managing context costs. Take the full course: claude.com/courses

May 9, 2026·3m·Watch on YouTube ↗

📖 CHAPTERS

1. 1

   What MCP is and why Claude Code needs it

   The video defines Model Context Protocol (MCP) as an open standard that connects Claude Code to external tools and data sources. The core motivation is that important “context” often lives outside your editor—so MCP helps Claude fetch what it needs at the right time.

2. 2

   Tools in agentic AI: actions, not just text output

   This segment explains the concept of “tools” in agentic AI: they enable the model to take actions to complete tasks, not merely respond with text. MCP is positioned as the mechanism that provides Claude Code with access to these tools.

3. 3

   Concrete MCP examples: Linear issues and dependency documentation

   The video gives examples of MCP servers that connect to specific sources of truth. A Linear MCP server can pull issue details, while another server can provide up-to-date dependency documentation to support development work.

4. 4

   Finding connectors: the MCP ecosystem

   This chapter points to the broader MCP connector ecosystem available online. It highlights that there are many prebuilt connectors available for common tools and services.

5. 5

   Adding MCP servers in Claude Code

   The workflow for configuring MCP is introduced, centered on adding servers via a command. This enables Claude Code to start using external services as tools.

6. 6

   Two server types: HTTP (remote) vs STDIO (local)

   The video distinguishes between remote MCP servers accessed over the network and local MCP servers that run as processes on your machine. Choosing between them depends on whether the service is hosted externally or executed locally.

7. 7

   Managing connections during a session with /MCP

   This section describes how to inspect and manage MCP servers from inside a Claude Code session. You can view what’s connected, check status, and disable servers you don’t want active.

8. 8

   Scoping servers for individuals vs teams (local, user, project)

   The chapter explains three scoping options for MCP servers, including a team-friendly approach. Project scoping uses a checked-in configuration file so everyone on the repo gets consistent tool access.

9. 9

   Context window cost: tool definitions consume tokens

   A key caution: MCP servers add tool definitions to the context window even if you’re not actively using them. Too many configured servers can reduce the remaining context available for your actual problem and code.

10. 10

    Context-efficient alternatives: CLIs and Skills

    The video recommends alternatives when context becomes constrained. If a tool has an equivalent CLI (e.g., GitHub or AWS), using the CLI can be more context efficient; “Skills” can also load on demand by name/description.

11. 11

    Tool search mode: what happens when tools exceed 10% of context

    This chapter explains Claude Code’s behavior when MCP tools consume too much of the context window. In that case, it switches to “tool search mode,” discovering tools on demand—though it may be less reliable than having tools already in context.

12. 12

    Practical recap: setup, scoping, and keeping context healthy

    The closing recap reinforces what MCP does, how to add servers, and why scoping matters for teams. It ends with operational advice: scope via `.mcp.json` for shared setups and disable unused servers to protect context budget.