Claude Context Window Exceeded: How to Fix It (2026)

When Claude Code shows a "context window exceeded" error, the fix is to start a new conversation or run /clear to wipe the context. The 200,000-token context window fills up silently during long sessions — and once it's full, Claude stops responding until you free space. Here's how to handle it before it kills your flow.

• Claude's context window is 200,000 tokens (roughly 150,000 words or ~500KB of code)
• Long agentic runs, large file reads, and repeated tool calls are the main culprits
• Usagebar tracks your context usage in the macOS menu bar and alerts you at 75% and 90% so you can act before hitting the wall

What does "context window exceeded" mean in Claude Code?

Every conversation with Claude accumulates tokens: your prompts, Claude's replies, file contents, tool outputs, and system instructions all count. When the running total exceeds 200,000 tokens, Claude Code cannot accept new input and returns a context window error. The session's history is too large to process.

This is distinct from a rate limit error (which is about requests per minute) or a message limit error (which is about your plan's usage cap). Context window errors are about the size of a single conversation, not your account quota.

How do you fix the Claude context window exceeded error?

Run /clear in Claude Code to wipe the conversation history and reset the context window to zero. This is the fastest fix. You can also start a completely new session, which achieves the same result. Neither action counts against your usage limits — you're just discarding history.

Option 1: use /clear

Type /clear at any point in your Claude Code session. This wipes all conversation history from the current window. Claude loses context of what you were working on, so you'll need to re-orient it briefly. This is the recommended approach for most developers.

Option 2: use /compact

Claude Code includes a /compact command that summarizes conversation history instead of deleting it. This compresses older turns into a shorter summary, reclaiming token headroom while preserving some context. It's useful when you're mid-task and don't want to re-explain the full codebase. According to the Claude Code slash commands docs, /compact accepts an optional custom summarization prompt if you want to guide what's retained.

Option 3: start a fresh session

Close the current Claude Code session and open a new one. Pass a concise project brief in your first message rather than relying on accumulated history. Most engineers working on large codebases do this naturally every few hours anyway.

Option 4: reduce what goes into context

Before hitting the limit again, change your workflow to use less context per session. See the section on prevention below.

Why does the context window fill up so fast?

A 200,000-token limit sounds enormous but burns through quickly in agentic sessions. Each tool call — reading a file, running a bash command, searching the codebase — adds both the request and the output to the context. A single large file read can consume 5,000-20,000 tokens. A multi-step refactoring task across a mid-sized codebase can exhaust the full window in one session.

Common culprits:

• Reading large files in full — Claude reads entire files even when only a few lines are relevant
• Long agentic runs — each tool call appends its output to the history
• Repeated back-and-forth — long exploratory conversations before any code is written
• Verbose error output — stack traces and test runner output can be thousands of tokens each
• System prompts and CLAUDE.md — project-level instructions are injected into every turn

For a deeper look at why token usage spikes, see why does Claude Code use so many tokens.

How do you prevent context window errors in Claude Code?

The most reliable prevention strategy is to run /compact proactively every 30-60 minutes during long sessions, before you hit the limit. Combine this with a leaner workflow that avoids dumping large files into context unnecessarily.

Practice	Token savings	Notes
Run /compact every hour	High	Summarizes old turns, preserves intent
Ask Claude to read specific line ranges	Medium	Avoids reading 1,000-line files for a 10-line fix
Keep CLAUDE.md concise	Low-Medium	Injected on every turn; trim ruthlessly
Split work across sessions by feature	High	Start fresh for each new task area
Avoid pasting raw error logs	Medium	Paste only the relevant lines

More detailed tactics are covered in how to reduce Claude Code token usage.

How do you monitor context window usage in real time?

Claude Code does not show a running token count in the UI by default. You can run /usage to see a snapshot of your current session usage, or check claude.ai/settings/usage for account-level data. Neither gives you a persistent, real-time view as you work.

Usagebar fills this gap. It's a $9 one-time purchase that sits in your macOS menu bar and shows your Claude Code context usage, 5-hour window progress, and weekly cap in real time. It sends notifications at 50%, 75%, and 90% thresholds so you can run /compact before the session breaks — not after.

This is especially useful for long agentic runs where you're not watching the terminal closely. You'll see the context gauge climbing in your menu bar and can take action on your own schedule.

See also: how to check Claude Code usage limits and Claude Code context window explained.

Does clearing context reset your usage limits?

No. Running /clear or starting a new session resets the conversation history for context window purposes, but it does not reset your plan's usage limits. Your 5-hour token budget and weekly cap continue to count down regardless of how many sessions you start. Context window and usage limits are tracked independently.

For details on how usage limits work separately from context, see does Claude Code usage affect Pro limits and when does Claude Code usage reset.

Key takeaways

1. The "context window exceeded" error means your session hit the 200,000-token limit — not an account quota issue.
2. Run /clear to reset immediately, or /compact to summarize and preserve partial context.
3. Long agentic runs, large file reads, and verbose error output are the main causes.
4. Proactively compact every hour during heavy sessions to avoid hitting the wall mid-task.
5. Usagebar ($9 one-time) shows live context usage in your menu bar and warns you at 75% and 90% before the session breaks.

Sources

• Claude Code slash commands documentation
• Using Claude Code with your Pro or Max plan — Anthropic Support
• Anthropic developer documentation