[BUG] Claude claims it doesn't have a tool, then immediately uses it in the next response

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?

During a Claude Code session, Claude incorrectly claimed that it did not have access to a specific tool or capability. When asked again (or in the very next response), Claude successfully used that exact tool without any issue — as if it had known about it all along.

This is inconsistent and confusing behavior: Claude should either have the tool or not. Claiming ignorance of a tool and then immediately using it suggests a context/state awareness bug where Claude's knowledge of its own available tools is unreliable within a session.

What Should Happen?

Claude should consistently know which tools are available to it throughout the entire session. If a tool is available, Claude should use it without first denying its existence. There should be no situation where Claude says "I don't have that tool" and then proceeds to use it in the same or next response.

Error Messages/Logs

Steps to Reproduce

1. Start a Claude Code session.
2. Ask Claude to perform a task that requires an mpc tool.
3. Claude responds saying it does not have access to that tool or that the capability doesn't exist.
4. Ask Claude the same question again (or continue the conversation naturally).
5. Claude now uses the exact tool it previously denied having, and completes the task successfully.

See attached screenshot for a concrete example of this happening.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.92 (Claude Code)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Terminal.app (macOS)

Additional Information

Screenshot showing Claude first denying it has the tool, then immediately using it in the follow-up response:

!Image