A recurring complaint with MCP is context bloat. If you have 10 MCP servers (not unheard of) and each one has an average of 5 tools, that’s 50 tool definitions that your model has loaded in its context, in addition to its own tools. In the case of a coding agent, these can be numerous. So you’re losing context, arguably penalizing the performance of your model and paying more for your LLM usage, even though you end up using only a very small subset of the MCP tools in practice.
this is where I’m at presently. But I do like the GPT -> Figma integration. But then again I fell back to mermaid and ascii charts for brainstorming.
Quote Citation: Moncef Abboud, “Is MCP Worth the Hype? | Moncef Abboud”, 2025-12-15, https://cefboud.com/posts/is-mcp-overhyped/