MCP Standard Year in Review
https://blog.modelcontextprotocol.io/posts/2025-11-25-first-mcp-anniversary/
The blog post describes how in one year, MCP transformed from a small open-source experiment into a de facto standard in the industry. Major companies like Notion, Stripe, GitHub, OpenAI, Microsoft, and Google have created their own servers to automate workflows. For centralized discovery and management of these servers, the MCP Registry was launched, becoming a single catalog for the entire ecosystem.

Coinciding with the anniversary, the team is releasing a new version of the MCP specification (November 2025). Key innovations include support for task-based workflows (for long-running operations), simplified and more secure authorization mechanisms, and an extensions system that allows adding specific functionality without changing the core protocol.

MCP Container Catalog
https://hub.docker.com/mcp
The site hosts a large library of ready-to-use, containerized MCP servers created by the developer community and powered by Docker technology. The servers are grouped by categories. The platform's goal is to simplify the use of MCP tools.

MCP Problems
https://www.youtube.com/watch?v=4h9EQwtKNQ8

The author argues that while MCP is a great idea in theory, in practice it has a serious problem that makes it ineffective. This problem is poor context management.

When an AI agent connects to MCP servers, all descriptions of available tools (tool definitions) are loaded into the language model's "context window." When the agent uses a tool, all results of its work (including intermediate data that may be unnecessary) are also sent to the context.

This consumes a huge number of tokens. The author gives an example where just two connected servers take up 20k tokens. It only gets worse with each iteration. The author calls this problem "context rot".

Agent Skills as an Alternative
https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview
https://www.youtube.com/watch?v=fOxC44g8vig

The solution once proposed by Cloudflare is that after finding the right tool, the agent generates code (e.g., TypeScript) to call this API. Based on this idea, Anthropic later proposed Agent Skills - a deep dive into the technology is available at https://leehanchung.github.io/blogs/2025/10/26/claude-skills-deep-dive/

#mcp #agentskills