Long-context strategies

Long-context strategies are prompting techniques and system-design patterns that help you get the best results when working with large amounts of text—typically 20,000 tokens or more—within Claude's context window. The context window is everything Claude can 'see' at once: your system prompt, conversation history, uploaded documents, tool results, and your current message. Getting the most out of that space requires deliberate choices about what to include, where to place it, and how to manage it over time. The core strategies fall into three categories: (1) structural placement—putting long documents near the top of your prompt and your actual question at the bottom, (2) document formatting—wrapping content in XML tags so Claude can navigate it clearly, and (3) context lifecycle management—using tools like server-side compaction and prompt caching to keep conversations coherent and cost-effective over many turns or very long sessions. These are not features you enable with a toggle. They are techniques you apply when writing prompts or building applications. They work on every Claude plan and across every interface—Claude.ai chat, Claude Code, and the API—though some advanced capabilities like server-side compaction require specific API headers or are rolling out in beta.