Claude Code hit a wall:

Error: Context length exceeded. Maximum context length is 200,000 tokens.

Or responses are getting worse as context grows. Here is how to fix it.

Fix 1: Use /compact regularly

Claude Code’s built-in context compressor:

/compact

This summarizes the conversation and frees up context space. Use it when:

• Context feels sluggish
• You get “context length” errors
• After completing a major task

Fix 2: Select files manually

Instead of letting Claude read everything, specify files:

# Add specific files to context
/add src/auth/login.ts
/add src/api/routes.ts

# Remove files from context
/drop src/old-module.ts

Only load files relevant to the current task.

Fix 3: Use smaller models for exploration

Use Sonnet for initial exploration, Opus for implementation:

/model claude-sonnet-5
# Explore the codebase, understand structure

/model claude-opus-4-8
# Implement the changes

Fix 4: Break tasks into smaller pieces

Instead of one large task, split it:

# Bad: Too much context
claude "refactor the entire auth system"

# Good: Focused tasks
claude "add password validation to login.ts"
claude "add rate limiting to API endpoints"
claude "update tests for auth changes"

Fix 5: Use git context

Claude Code can read git history efficiently:

# Ask about recent changes
claude "what changed in the last 5 commits?"

# Ask about specific files
claude "explain the changes in src/auth/login.ts"

Git context is more efficient than reading entire files.

Fix 6: Exclude large files

Add to .claudeignore:

# .claudeignore
node_modules/
dist/
*.min.js
large-data.json
test/fixtures/

Claude Code will skip these files during context building.

Fix 7: Use CLAUDE.md for persistent context

Put project overview in CLAUDE.md:

# CLAUDE.md

## Project structure
- src/auth/ - Authentication module
- src/api/ - REST API routes
- src/models/ - Database models

## Key decisions
- Using TypeScript strict mode
- PostgreSQL for database
- JWT for authentication

## Current task
Working on adding rate limiting to API endpoints.

This provides context without using conversation tokens.

Still overflowing?

For very large codebases:

1. Use repository map — Claude Code builds a map of your codebase
2. Focus on one module — Work on one part at a time
3. Use Aider — Better at handling large codebases with repo map
4. Split into microservices — Smaller codebases are easier to manage

FAQ

What is the context limit for Claude Code?

Claude Sonnet 5 has a 1M token context window. However, Claude Code’s effective limit is lower because it includes system prompts, tool definitions, and conversation history. In practice, you may hit limits around 200-300K tokens of actual content.

How do I know how many tokens I am using?

Claude Code does not show token counts directly. Monitor your API usage in the Anthropic console. If you are hitting limits frequently, use /compact to reduce context.

Can I increase the context limit?

Not directly. The context limit is set by the model. However, you can work around it by: using /compact regularly, selecting specific files with /add, and breaking tasks into smaller pieces.

Related: Claude Code Complete Guide · Aider Complete Guide · Context Window Management · Best AI Models for Large Codebases · Claude Code vs Aider