What happened
Anthropic introduced a new approach called "Code Execution with MCP" that transforms MCP agents into code-first systems. This method turns MCP tools into executable code functions, reducing the need to push every tool definition and intermediate result through the model context. The approach aims to improve efficiency by lowering token usage and latency in AI agents.
Why it matters
- It significantly reduces token consumption and latency for AI agents using MCP.
- The method enables more efficient chaining and composition of tools through code rather than context passing.
- It enhances AI agents' capabilities to handle complex programming tasks securely and reliably.
Key facts
- 1. The "Code Execution with MCP" approach was publicly detailed by Anthropic on November 4, 2025.
- 2. Token usage can be reduced by up to 98.7% using this code execution method compared to traditional MCP tool calls.
- 3. Anthropic released `mcp-server`, an open-source implementation that supports secure, sandboxed code execution for AI models.
