[BUG] - Repeated incorrect capability claims led to hours of unnecessary work at 3am

Status Open
Reported on v2.1.209
Maintainer reply None cached
Activity 0 comments · opened Jul 16, 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?

I am a paying Claude Code user. Tonight I asked Claude to send an email via Zoho Mail. Claude repeatedly told me this was impossible — that Zoho was read-only and could only search/list/read emails, not send them.

I told Claude multiple times that it could send via API and Bash, based on my own experience using it in Claude.ai chat. Claude overrode my corrections every time and insisted it couldn't be done, then spent hours building unnecessary local MCP servers to "solve" a problem that didn't exist.

At the end of the session Claude tested a direct Bash/API call and it worked immediately — exactly as I had been saying the whole time.

Impact:

Hours of session time wasted
Significant usage credits consumed building infrastructure that was never needed
Working at 3am on something completely unnecessary
Claude ignored repeated user corrections and pushed its own incorrect assumption

What Should Happen?

Test the direct approach first. When the user says something works, believe them. I am requesting a review of this session and consideration of usage credit reimbursement as this has happened several times with Code and Chat, more with Chat than code. I have never thought to report the issue until now...

Error Messages/Logs

Steps to Reproduce

  1. Ask Claude Code to send an email via Zoho Mail
  2. Claude claims Zoho is read-only and cannot send
  3. Tell Claude it can send via API/Bash based on prior experience
  4. Claude overrides the correction and insists it cannot send
  5. Claude spends hours building unnecessary local MCP servers as a "workaround"
  6. Claude finally tests a direct Bash/API call — it works immediately

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.209

Platform

Other

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Using Claude Code Desktop App, not CLI.
Claude ignored repeated user corrections multiple times.
User stated Zoho send worked via API/Bash. Claude insisted
it was impossible. Hours of usage credits were consumed
building unnecessary MCP infrastructure. This is a
reasoning/listening failure, not a code bug. Requesting
review and consideration of credit reimbursement.

View original on GitHub ↗