Docker Just Fixed 90% of AI Coding By Releasing This

When MCP expands from a couple local servers to hundreds, AI coding starts failing in familiar ways: bloated context windows, wasted tokens, and tool results drowning the signal. This video breaks down Docker’s dynamic approach to MCP and how it’s meant to keep agents lightweight while still supporting more autonomous, tool-driven workflows.

Key takeaways

• Clarifies the MCP challenges Docker calls out: which servers to trust, how to avoid shipping unused tool definitions into context, and how agents can discover/configure tools efficiently.
• Shows Docker’s MCP catalog of verified servers and a setup where your MCP client connects to Docker while Docker manages your MCP servers.
• Explains the MCP gateway and tools like MCP find/add/remove for pulling in only the tools you need.
• Demonstrates “code mode,” where agents generate JavaScript-enabled tools that can call other MCP tools, run in a sandbox, and persist state via volumes.