Model Context Protocol

Extending Claude with Model Context Protocol (MCP)

The Model Context Protocol (MCP) is the bridge between Claude Code and your external ecosystem. By configuring MCP servers, you transform Claude from a local code assistant into a fully integrated teammate capable of interacting with your project management tools, databases, and communication platforms.

Through MCP, Claude can move beyond your local file system to read Jira tickets, post Slack updates, query production logs, or even trigger CI/CD pipelines—all through a unified interface.

The .mcp.json Configuration

Claude Code looks for a .mcp.json file in your project root to determine which external tools it should have access to. This file defines the connection strings and commands needed to spin up local or remote MCP servers.

Each entry in the configuration identifies a server that Claude can use to fetch data or execute actions.

{
  "mcpServers": {
    "linear": {
      "command": "npx",
      "args": ["-y", "@linear/mcp-server"],
      "env": {
        "LINEAR_API_KEY": "your_api_key_here"
      }
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "your_token_here"
      }
    }
  }
}

Key Integrations

The most powerful applications of MCP involve connecting your development workflow to your planning tools:

• Project Management (Jira/Linear): Claude can read the description of a ticket, identify the acceptance criteria, write the code, and then update the ticket status to "In Review."
• Version Control (GitHub/GitLab): Claude can search across repositories, read pull request comments, and help resolve merge conflicts.
• Documentation (Confluence/Notion): Claude can pull in external design specs or product requirements documents to ensure the code it writes aligns with business goals.

Example Workflow: The /ticket Command

With MCP configured, you can create custom commands (like .claude/commands/ticket.md) that leverage these external connections to automate entire cycles of work.

How it works in practice:

1. Read: You type /ticket PROJ-123.
2. Context: Claude uses the MCP server to fetch ticket details from your project management tool.
3. Plan: Claude analyzes the requirements against the local codebase.
4. Execute: Claude implements the requested changes.
5. Verify: Claude runs the local test suite.
6. Report: Claude posts a summary back to the ticket and prepares a pull request.

Adding New Tools

To add a new capability to Claude Code:

1. Identify an MCP Server: Find a pre-built server for your tool (e.g., Slack, Postgres, Google Drive) or build a custom one.
2. Update .mcp.json: Add the server configuration to your project.
3. Restart Claude Code: Claude will automatically detect the new tools and describe them to you upon startup.

Once connected, you don't need to use specific syntax for most interactions. You can simply tell Claude: "Check the latest comments on the 'User Auth' ticket in Jira and update the login logic to match," and Claude will use the MCP bridge to handle the rest.