[BUG] Pro login loads 2.9M tokens in "System tools" — makes Claude Code unusable
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?
When logged in via Claude.ai Pro OAuth (/login → "Claude.ai Subscription"), Claude Code loads approximately 2.9M tokens into "System tools" on startup, making the context 1460% full before any user input.
Steps to reproduce:
Log out: rm ~/.claude/.credentials.json
Start claude in any project directory → /context shows ~721 tokens (normal)
Run /login → authenticate with Claude.ai Pro subscription
/context immediately shows 2.9M tokens in "System tools" (1460% full)
Environment:
Claude Code version: 2.1.133
Platform: Ubuntu 24.04 LTS, linux-x64
Model: claude-sonnet-4-6
Plan: Claude Pro
Interface: Terminal + VS Code extension
What Should Happen?
Expected behavior: System tools should load ~10-15k tokens (built-in tools only), regardless of login method.
Actual behavior: Pro OAuth login causes 2.9M tokens to load in "System tools". No MCP connectors are connected (verified in claude.ai → Settings → Connectors).
Error Messages/Logs
Steps to Reproduce
see description
Claude Model
Sonnet (default)
Is this a regression?
I don't know
Last Working Version
_No response_
Claude Code Version
2.1.133
Platform
Anthropic API
Operating System
Ubuntu/Debian Linux
Terminal/Shell
Other
Additional Information
OAuth token scopes include user:mcp_servers — this may be triggering automatic MCP server loading
Removing .credentials.json and using API key only resolves the issue
No MCP connectors visible in claude.ai settings
6 Comments
Found 3 possible duplicate issues:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
ENABLE_CLAUDEAI_MCP_SERVERS=false should disable the claude.ai connector sync while keeping OAuth login. Anthropic added the flag in v2.1.14 (see #50062). If setting it in your shell drops /context back to roughly 10-15k tokens, then the 2.9M is account-level connector inheritance. If /context still shows 2.9M with the flag set, the v2.1.14 fix may not be covering whatever path is firing in v2.1.133.
Curious which way it goes. The result tells whether this is a duplicate of #50062 or a separate regression.
Hi yurukusa!
Setting ENABLE_CLAUDEAI_MCP_SERVERS=false in ~/.claude/settings.json env reduces context to 721 tokens when not logged in via OAuth. However, after /login with Claude.ai Pro subscription, System tools still shows 2.9M tokens — the flag does not suppress MCP server loading when authenticated via OAuth in v2.1.133.
https://github.com/anthropics/claude-code/issues/50062 has been closed and is not editable for me. My message is: The proble remains still in the current Version.
Duplicate to #20412
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.