MCP: Model Context Protocol

Real engineering work usually touches more than just a codebase — issue trackers, internal databases, design tools, documentation systems. MCP (Model Context Protocol) is an open standard for connecting an AI agent to systems like that, so each integration is built once, as a standard MCP server, instead of as a bespoke connector for every tool an agent might ever need to reach.

The problem it solves

Without a shared protocol, connecting an agent to a new external system means building a custom integration for that specific system, every time. MCP fixes the "every time" part: any tool that speaks MCP can connect to any MCP-compatible agent, including Claude Code, without custom glue code on either side. Anthropic and the broader community maintain MCP servers for many common systems already, and writing one for an internal tool follows the same standard interface.

What connecting one looks like

Adding an MCP server to Claude Code means pointing it at the server (a URL or a local process, depending on the server) and, where required, providing credentials for it to authenticate with the underlying system. Once connected, whatever capabilities that server exposes — querying an issue tracker, searching a knowledge base, reading from a database — become available as tools, the same way built-in tools like file editing and search are.

Project-level vs personal connections

Like custom commands, MCP servers can be configured at the project level (shared with everyone working on the codebase, checked into project configuration) or personally (your own connections, available across whatever projects you work on). A project's issue tracker integration probably belongs at the project level, shared with the team; a personal note-taking tool you use across many unrelated projects belongs in your own configuration.

Representative use cases

• Issue trackers — reading ticket details, updating status, or creating new tickets as part of a workflow that starts from a bug report and ends with a linked fix.
• Design tools — pulling specifications or assets referenced by a ticket directly into context rather than describing them by hand.
• Internal databases — querying real data to verify a fix or understand the shape of a problem, without writing one-off scripts just to look something up.
• Documentation systems — searching internal docs that, unlike public documentation, web search has no way to reach.

Why this matters more as the tool ecosystem grows

The value of a standard protocol compounds with the number of tools that adopt it: a single MCP connection in Claude Code can reach any MCP server, and a single MCP server you build for your organization's systems can be used by any MCP-compatible client, not just Claude Code specifically.

Next: running something long in the background without blocking the rest of the session.