Claude Code 429 Error: Why You Hit Rate Limits With Credits Left
The TPM Problem
API rate limits have two dimensions: requests per minute (RPM) and tokens per minute (TPM). Even with a full credit balance, you can't send 200,000 tokens every 30 seconds. Claude Code's agentic behavior — scanning files, reading git logs, checking package.json — consumes 'hidden' tokens before you even ask it to write code.
Context Bloat
By default, Claude Code sends your entire conversation history with every request. After a long coding session, this can be 100,000+ tokens per message. The fix: run /compact inside Claude Code to compress your conversation history. Run it every 50 messages or whenever context exceeds 20,000 tokens.
The .claudeignore Fix
Create a .claudeignore file in your project root (like .gitignore) to exclude build artifacts, node_modules, dist folders, and large data files. This prevents Claude from scanning everything and burning tokens on irrelevant files.
Use the --include Flag
Scope Claude to only the directories that matter: claude --include src/. This prevents full-repo scans that exhaust your TPM before you even start coding.
Upgrade Your Tier
Anthropic's rate limits scale by usage tier. Tier 1 allows roughly 50 RPM; higher tiers scale to 4,000 RPM. If you're hitting limits regularly, upgrading your tier or switching to Claude Max ($100-200/month) gives you significantly higher limits at a fixed monthly cost.
Frequently Asked Questions
Why do I get 429 errors with credits remaining?
Rate limits are per-minute, not per-month. You can have a full credit balance but hit the tokens-per-minute ceiling during intensive coding sessions.
What's the best way to prevent 429 errors?
Use /compact regularly, create a .claudeignore file, scope with --include, and consider upgrading to Claude Max for higher limits.
Skip the guesswork. Get DevLaunch.
Interactive setup wizard + AI debugger for MiroFish, OpenClaw, and Claude Code.
GET DEVLAUNCH — $27 →