Claude Code Context Window: What Each Category Means and How to Manage It

Claude Desktop now shows a live context window breakdown directly in the session interface — no terminal command needed. You can see exactly how full your context window is, which categories consume the most tokens, and how much free space remains. This guide explains what each category in the breakdown means and what to do when the numbers start climbing.

This post documents the breakdown panel: what triggers it, how to interpret each row, and four practical actions you can take during a long session to extend your working budget.

What is the context window?

The context window is Claude’s working memory for a session. Every message you type, every response generated, every file read, and every tool called occupies space in a fixed budget measured in tokens. A token is roughly four characters of English text. A single research paper or grant document can consume tens of thousands of tokens — a long afternoon session accumulates even faster.

The standard Claude Code context window is 200,000 tokens. That sounds substantial, but it fills quickly when you combine conversation history, loaded tools, custom instructions, and document contents. When it approaches capacity, Claude begins compressing earlier parts of the conversation — and that is when response quality starts to slip.

How to access the breakdown

In any active Claude Code session running in Claude Desktop, look for the context window indicator at the top of the session. It shows your current usage as a fraction and percentage — for example, 161.3k / 200.0k (81%). Click it to open the full breakdown panel with a category-by-category token count.

Alternatively, you can run the /context command in the terminal to see a similar breakdown with additional optimisation suggestions.

How to read the breakdown: what each category means

Here is what a real breakdown looks like mid-session — at 161,300 tokens used out of a 200,000-token window (81% full). Each category below corresponds to what you see in the panel.

Messages — conversation history

What it is: Everything you have typed and everything Claude has replied — the full conversation history.

Why it matters: This is almost always the largest category and the one that grows fastest during a long session. You cannot reduce what is already there, but you can stop it compounding by running /compact before the window fills.

Typical percentage: 60–80% in extended sessions.

System prompt — core operating instructions

What it is: Claude Code’s core operating instructions — loaded automatically at the start of every session.

Why it matters: This overhead is fixed and cannot be reduced. It defines how Claude behaves, what tools it has access to, and how it approaches coding tasks.

Typical percentage: 1–3%.

Skills — custom instructions

What it is: Custom instructions you have loaded — for example, a CLAUDE.md file in your project folder.

Why it matters: This category is optimisable. Trimming redundant instructions or splitting a large CLAUDE.md into smaller, session-specific files reduces it directly.

Typical percentage: 1–5%, depending on how many custom instruction files you’ve loaded.

MCP tools — actively used integrations

What it is: Tool definitions for the MCP integrations you have actively used in this session — Google Drive, Canva, MailerLite, and others. Each connected integration adds its tool schema to the budget the moment it is called.

Why it matters: This shows you which tools are currently loaded and contributing to your token usage.

Typical percentage: 0.5–3%, depending on how many tools you’ve invoked.

MCP tools (deferred) — connected but unused

What it is: Tool definitions for every MCP integration you have connected but not yet called in this session. Claude loads them defensively so they are ready when needed — but they consume budget whether or not you use them.

Why it matters: This is often the single largest optimisation available. If you have integrations connected that you will not need in the current session, disconnecting them frees this space immediately.

Typical percentage: 20–50% in sessions with many connected MCP servers. This can be the biggest contributor to context usage before you even start working.

System tools (deferred) — built-in tools

What it is: Built-in Claude Code tools (file operations, bash commands, search) that are loaded but not yet invoked. Like deferred MCP tools, these sit ready in the background.

Why it matters: This overhead is fixed and cannot be reduced.

Typical percentage: 3–6%.

Autocompact buffer — reserved space

What it is: Reserved space — approximately 13,000 tokens — held back so Claude can finish generating its current response when auto-compaction triggers.

Why it matters: Most users never knew this buffer existed. It explains why sessions seem to compact slightly before the window is technically full. This overhead is fixed and not optimisable.

Typical percentage: 15–20%.

Free space — remaining budget

What it is: Remaining budget before compaction triggers. At 0%, the window is at capacity.

Why it matters: Seeing this reach single digits is your cue to act — either compact manually or disconnect unused integrations.

Typical percentage: Varies throughout the session. Ideally, you want at least 20–30% free when starting complex tasks.

Four strategies to extend your context budget

Once you can see where your tokens are going, you can act on it. Here are four approaches ordered from quickest to most structural.

1. Disconnect unused MCP integrations

If the “MCP tools (deferred)” category is consuming 30–50% of your window, and you recognise integrations you will not need in this session, disconnect them. This frees space immediately without losing any work.

How to do it: Open Claude Desktop settings, navigate to the MCP integrations panel, and toggle off the servers you are not using. The change takes effect instantly.

2. Run /compact manually before the window fills

Auto-compaction triggers when the window reaches capacity, but you can run it proactively. This compresses the oldest parts of the conversation into a summary and frees up space to continue.

How to do it: Type /compact in the Claude Code session and press Enter. Claude will summarise the conversation history and resume with a clean slate.

When to use it: Before starting a new, complex task when your context usage is already above 70%.

3. Split large CLAUDE.md files into session-specific instructions

If your “Skills” category is consuming 5–10% of your window, and you have a single large CLAUDE.md file with instructions for multiple workflows, split it into smaller files. Load only the instructions relevant to the current session.

How to do it: Create separate CLAUDE.md files for different project types or workflows. Place them in subdirectories and load them selectively using path-scoped rules.

4. Start a new session for unrelated tasks

If you have finished one task and are about to start a completely unrelated one, exit the current session and start fresh. This gives you a clean context window without carrying over irrelevant conversation history.

How to do it: Exit the current session (Ctrl+C or type exit) and start a new one with claude.

What happens when the window fills: auto-compaction

If you reach the limit without intervening, Claude Code triggers auto-compaction automatically. This compresses the oldest parts of the conversation into a summary and frees up space to continue. The threshold is set at the total window size minus the Autocompact Buffer shown in the breakdown.

If auto-compaction fails three consecutive times, the circuit breaker stops it and you will need to run /compact manually.

Using the context breakdown regularly means you will see the window filling in time to compact deliberately — on your schedule, not Claude’s.

When to consider upgrading your context window

If you regularly hit the 200,000-token limit despite active management, the Max, Team, and Enterprise plans for Claude Code support a 1,000,000-token context window via Opus 4.6 and Opus 4.7. This removes the constraint entirely for most long-form research and writing workflows.

The upgrade is worth evaluating if you work with large document sets, run extended multi-phase sessions, or use several MCP integrations simultaneously.

Pro plan users can access the 1M context window by enabling extra usage — type /extra-usage in Claude Code to unlock it.

Related reading on The Science Talk

This guide accompanies the full context window breakdown post on The Science Talk — which includes a detailed walkthrough of each category with visual examples and real-world session data.

Want more guides like this? Browse all AI Guides or get in touch →