Preflight Checklist
What's Wrong?
Since today (Feb 13), Claude Desktop has not been playing nicely with our local MCP server that has been working for the past 3-4 months. The model is still the same, Sonnet 4.5. This is tested on a simple query that has always worked before.
- Doesn't understand it should use the MCP server unless explicitly stated to do so, then
Searched available tools is done and it finally loads the available tools. This step has never happened before.
- Doesn't seem to read the full tool instruction, because it doesn't provide mandatory arguments into a tool call. It also provides the wrong format into an argument (eg, it provides a
list[str] into an argument that expects list[list[str]] type).
Note that this same query is still working on our langchain agent that uses langchain-anthropic.
What Should Happen?
As described in the problem, it should understand tool descriptions and arguments because the same MCP server has always worked with the same test query on Claude Desktop until today.
Error Messages/Logs
Steps to Reproduce
Our MCP server code is proprietary.
Claude Desktop version is Claude 1.1.3189 (1b7b58)
Model is Sonnet 4.5
Since the Searched available tools step seems to be new, I'm assuming it's not able to grab the full context of tool descriptions or something?
Claude Model
Sonnet (default)
Is this a regression?
Yes, this worked in a previous version
Last Working Version
No response
Claude Code Version
1.1.3189
Platform
Other
Operating System
Windows
Terminal/Shell
Other
Additional Information
No response
Preflight Checklist
What's Wrong?
Since today (Feb 13), Claude Desktop has not been playing nicely with our local MCP server that has been working for the past 3-4 months. The model is still the same, Sonnet 4.5. This is tested on a simple query that has always worked before.
Searched available toolsis done and it finally loads the available tools. This step has never happened before.list[str]into an argument that expectslist[list[str]]type).Note that this same query is still working on our langchain agent that uses
langchain-anthropic.What Should Happen?
As described in the problem, it should understand tool descriptions and arguments because the same MCP server has always worked with the same test query on Claude Desktop until today.
Error Messages/Logs
Steps to Reproduce
Our MCP server code is proprietary.
Claude Desktop version is Claude 1.1.3189 (1b7b58)
Model is Sonnet 4.5
Since the
Searched available toolsstep seems to be new, I'm assuming it's not able to grab the full context of tool descriptions or something?Claude Model
Sonnet (default)
Is this a regression?
Yes, this worked in a previous version
Last Working Version
No response
Claude Code Version
1.1.3189
Platform
Other
Operating System
Windows
Terminal/Shell
Other
Additional Information
No response