[BUG] Large MCP tools context warning - MCP tools and excessive context usage

[t34wrj]

Preflight Checklist

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

What's Wrong?

Despite no changes to my enabled MCP tools, I am now unable to run any commands without generating a 'Context low · Run /compact to compact & continue' warning.

When running the /doctor command, it shows the token usage of my MCP tools as:

Context Usage Warnings
└ ⚠ Large MCP tools context (~144,802 tokens > 25,000)
└ MCP servers:
└ MCP_DOCKER: 135 tools (~125,964 tokens)
└ chrome-devtools: 26 tools (~17,512 tokens)
└ grep: 1 tools (~1,326 tokens)

This was not an issue for me last week and I have not changed the MCP tools that are enabled.

What Should Happen?

Claude Code should be able to load the MCP tools without the extreme level of token usage.

Error Messages/Logs

Steps to Reproduce

Run any command.

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Unsure.

Claude Code Version

2.0.50

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

VS Code integrated terminal

Additional Information

No response

[github-actions]

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

[Calebperez23]

48h Long-Horizon A/B Validation Offer — Real MCP Token/Tool Cost Metrics

I’ve been working on long-horizon validation of tool cost behavior in MCP-linked agent workflows and found that short or immediate token counts don’t tell the full story in real usage.

I’m offering a 48-hour A/B validation sprint specifically on a workflow like the one described here — tool-heavy, long horizon, real workflows — where:
• pinned baselines are used,
• a stabilization window is included,
• ratios for tokens per interaction, p95 latency, and tool overhead are reported.

I’ll publish the methodology overview and results publicly afterward; implementation specifics remain private until a mutual evaluation agreement.

If you’d like a 20-minute sync + a 48-hour sprint to gather comparative metrics on your workload, reply here or DM.

— Caleb

[github-actions]

This issue has been automatically closed due to 60 days of inactivity. If you're still experiencing this issue, please open a new issue with updated information.

[github-actions]

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.