[BUG] every input/output takes 10% of my 5hr rate limit quota!!!

Resolved 💬 3 comments Opened Mar 23, 2026 by Raiy-TW Closed Mar 26, 2026

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

In one specific project, every input/output interaction consistently consumes approximately 10% of my 5-hour usage quota, regardless of the size or complexity of the request.

This behavior is abnormal and only occurs in this particular project. Other projects function normally with expected token usage.

Additional issues observed in this project:

  • Context usage is unusually high.
  • Automatic context compaction does not occur.
  • Manual compaction does not reduce context usage.
  • Disabling the MCP server does not resolve the issue.

As a result, the quota is exhausted extremely quickly, preventing me from continuing to use Claude Code for development work.

I am currently unable to identify any specific factor that could be causing this abnormal behavior.

What Should Happen?

  • Token usage should scale proportionally with the size and complexity of input/output.
  • Context should be automatically compacted when reaching limits.
  • Manual compaction should effectively reduce context size.
  • Disabling MCP server should not have no effect if it is unrelated.
  • Resource consumption should be consistent across different projects under similar conditions.

Error Messages/Logs

Steps to Reproduce

  1. Open the affected project (specific project where the issue occurs).
  2. Send a simple input (e.g., a short prompt or small code-related request).
  3. Observe the usage consumption after the response.
  4. Repeat a few interactions.

Observed result:

  • Each interaction consumes approximately 10% of the 5-hour quota.
  1. Attempt to manually trigger context compaction.

Observed result:

  • No noticeable reduction in context usage.
  1. Disable MCP server and repeat the same interactions.

Observed result:
The issue persists with the same abnormal usage.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

v2.1.81

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗