Context window
The context window is the amount of text a language model can consider at once, including the prompt, the conversation and tool results.
What it means
A model reads everything inside its context window before it writes. That includes the user's instructions, earlier messages, tool definitions and every tool result. When the window fills, older material is trimmed or the model loses track of it. Large tool results therefore compete with the work the user actually asked for.
Search MCP servers differ in how much text they return. A tool that returns a short answer with a few quotes leaves room for reasoning. A tool that returns full pages for every result can fill the window in a few calls. When you compare servers, look at the size of a typical result on your own questions, not just the list of features. Keeping results short is therefore a practical habit, because a model that spends its window on raw page text has less room left for the reasoning that produces a useful answer.
Continue this reading path
Cite this entry
https://searchmcpserver.com/glossary/context-window/