Limitations of Model Composition Protocol (MCP) and the Power of Code
By
Bogdanp
An everything bagel for the brain. Substantive, layered, well-seasoned.
Summary
The article discusses the limitations of the MCP (Model Composition Protocol) and emphasizes the importance of code in solving agentic flows. It highlights the challenges of composition and context demands in using MCP for tasks like completing a GitHub task.
Key quotes
· 2 pulledIt isn’t truly composable. Most composition happens through inference.
It demands too much context. You must supply significant upfront input, and every tool invocation consumes even more context than simply writing and running code.
You might also wanna read
Web-to-MCP: Send Website Components Directly to AI Coding Tools via Model Context Protocol
Web-to-MCP is a tool that bridges the design-to-development gap by allowing users to send pixel-perfect website components directly to AI co

How AI Coding Agents Are Reshaping Software Development: Capabilities, Limitations, and the Future
This article is the fourth in a series examining how AI is changing software development. The author reflects on the evolving capabilities o
Building a Zero-Dependency Python MCP Server for Local AI File Access
A developer describes building a lightweight, zero-dependency MCP (Model Context Protocol) server in pure Python to solve the pain point of

AI Companies Collaborate on Model Context Protocol for Interoperable AI Agents
Major AI companies including Anthropic, OpenAI, Google, and Block are collaborating on the Model Context Protocol (MCP), a standardized fram
Figma Releases Remote MCP for Live Design Integration with IDEs and AI Agents
Figma has released a new MCP (Model Context Protocol) that enables remote access to live design context from Figma files directly into IDEs
Exotel Launches MCP Server for AI Agents to Trigger Communications Without Coding
Exotel has launched a Model Context Protocol (MCP) Server that enables AI agents to trigger communication actions like making calls and send
