anthropics / anthropics/claude-code
Weekly usage limit ~30% consumed by Tuesday on a single MCP-heavy session — limits too tight for agentic CAD workflows
- Dominant language
- Python
- Stars
- 145k
- Forks
- 23.1k
- PR merge metrics
- PR metrics pending
Description
### Preflight Checklist
- [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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?
Environment: Claude Code desktop app (Windows 11), Sonnet 5 for most of the session, switched to Opus 5 at the end. Custom MCP bridge to Autodesk Revit 2026 (IronPython code execution tool).
What happened: One working session (a few hours) editing a roof element in a Revit model — roughly 80 MCP tool calls, several read-only geometry checks per edit, two full rebuilds of one element. By Tuesday afternoon the weekly limit shows ~30% used. Nothing exotic: no web browsing, no large repo indexing, one ~800-line script read once.
Why this matters: Agentic CAD/BIM work is inherently call-heavy: every edit needs a verification round-trip through the API, and tool results (element lists, coordinates) are the actual working data, not noise. With the reduced weekly limits this workflow burns through a week's budget in ~2 days, which makes the product unusable for its best use case.
Acknowledged on our side: part of the burn was avoidable (verbose list dumps, repeated queries). We've since added a helper prelude and stricter output rules. Even so, the remaining necessary volume does not fit the current weekly cap.
### What Should Happen?
Reconsider the weekly cap reduction, or
Don't count (or discount) tool-result tokens from local MCP servers the same as generated output, or
Expose a per-session token meter in the app so users can see the burn rate before the weekly warning.
### Error Messages/Logs
```shell
```
### Steps to Reproduce
na
### Claude Model
Sonnet (default)
### Is this a regression?
Yes, this worked in a previous version
### Last Working Version
_No response_
### Claude Code Version
2.1.221
### Platform
Anthropic API
### Operating System
Windows
### Terminal/Shell
Other
### Additional Information
_No response_
Contributor guide
No contributing guide indexed for this repository
Research direction
The report names the Claude Code desktop app, a custom MCP bridge to Autodesk Revit 2026, and version 2.1.221, but no repository files, tests, or reproducible steps. Start by examining how weekly usage limits and MCP tool-result accounting are exposed in the app; done would require an agreed cap or accounting change, or a per-session token meter.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- cloud, tooling
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100