[BUG] MCP file system not working
Claude Desktop MCP Complete Failure - Support Request
Problem Summary
I have been struggling with MCP (Model Context Protocol) functionality in Claude Desktop for an entire day. All local MCP servers are completely non-functional despite extensive troubleshooting efforts.
What I've Tried (Exhaustive Testing)
- ✅ Multiple servers/computers: Tested on different machines with fresh installations
- ✅ Multiple accounts: Tried different Claude accounts
- ✅ Complete reinstallation: Uninstalled and reinstalled Claude Desktop multiple times
- ✅ Configuration verification: Verified all config files are syntactically correct
- ✅ System reboot: Restarted computers multiple times
- ✅ Environment cleanup: Cleared all cache and temporary files
- ✅ Fresh setup: Created completely new configurations from scratch
Current Status
✅ Working:
- web_search tool
- web_fetch tool
- artifacts creation
❌ Completely Broken:
- All Windows-MCP tools (State-Tool, Powershell-Tool, Click-Tool, etc.)
- All filesystem MCP tools (read_file, list_directory, write_file, etc.)
- Analysis/REPL tool
Error Patterns
Primary Error: "Cannot convert undefined or null to object"
Secondary Error: "Error executing code: Cannot convert undefined or null to object"
Evidence This is NOT a Local Issue
- Multiple independent computers show identical failures
- Cloud-based MCP tools work perfectly (web_search, web_fetch)
- Only local MCP servers fail systematically
- Fresh installations on clean systems fail identically
Recent GitHub Evidence
I found GitHub issue #4188 reporting the exact same problem starting July 23, 2025:
- "Claude Desktop MCP tools failing with 'Required' parameter error"
- "Cannot convert undefined or null to object"
- Multiple users affected simultaneously
- Root cause identified: "Claude Desktop is not passing required parameters to MCP tools"
Timeline
- July 22, 2025: Everything worked perfectly
- July 23, 2025: Complete MCP failure began (today)
Technical Environment
- Claude Desktop Version: [Latest as of July 23, 2025]
- Operating System: Windows 11
- Node.js: Installed and verified working
- UV/Python: Properly configured
Impact & Request
I have spent an entire day troubleshooting this issue, following all documentation and trying every possible solution. This appears to be a widespread service issue affecting Claude Desktop's MCP parameter handling system.
Please:
- Confirm this is a known issue with Claude Desktop's MCP implementation
- Provide an estimated timeline for resolution
- Suggest any temporary workarounds if available
This has significantly impacted my productivity, and I would greatly appreciate acknowledgment that this is a service-side issue rather than user configuration problems.
Contact Information
- Support needed: Technical resolution or official acknowledgment
- Priority: High (complete loss of core Claude Desktop functionality)
Thank you for your time and assistance.
18 Comments
it was working all day for me and started to have exactly the same issue as of 1 hour ago.
any workarounds would be appreciated
Same issue on MacOS Sequoia 15.5, Claude Desktop 0.12.28
Same issue with Claude Code.
It crashes if there's a file modification.
I also had the same problem with Claude Desktop MCP until now.
I use Claude Teams.
The MCP tool I use is Filesystem Developed by [Anthropic].
yesterday everything worked. But today failure
Same problem both with filesystem and other MCP server.
Claude Desktop on Mac. Up to July 22 @4PM CET no problems.
Same. Claude Desktop error:
``
Error executing code: Cannot convert undefined or null to object``I also still had the versions app-0.12.20 back upped, it now has the the same problem
<img width="569" height="367" alt="Image" src="https://github.com/user-attachments/assets/8ffdb1bf-bac0-460d-9757-dcdfceb1229b" />
Also : claude desktop silently updated to 0.12.28 at 2AM in the morning, but was still working fine with the new version until 12:00 gmt+1 (around 2 hours ago)
Confirmed that Claude Desktop on windows and mac doesn't work.
Note, it's not jjust MCP. The new connectors all blow up with the screen:
Claude will return soon
Claude.ai is currently experiencing a temporary service disruption. We’re working on it, please check back soon.
Which is odd because the Status says operational....and it absolutely isn't.
When will it be fixed any updates please?
This has been broken all day, I can't use many MCPs now, I'm surprised there is no response from Anthropic yet, not even a "we are looking into it"
+1
I checked it just now still not working.
No annoucement from Anthropic also.
@ysystech it's 8:15 in the morning around San Francisco. The people with the knowledge to fix the issue probably just came to office around now
1.
When calling the filesystem as a Connector, the first connection is established normally, but it disconnects from the server in 0.3 seconds.
I've been struggling with this all day, uninstalling and reinstalling nodes, Cursors, and Claude AI. I've modified the index.js files countless times and tried using console.error for debugging, but nothing has worked (the console.error messages don't even appear in the log). I even deleted all the code in the wrapper's index.js file and replaced it with an error message like:
"
This is incorrect JavaScript code!!!
It can never be executed!!!
"
However, seeing the same usual logs appear as before, I realized the problem is beyond my control.
2025-07-23T13:16:27.561Z [Filesystem] [info] Client transport closed { metadata: undefined }
2025-07-23T13:16:27.561Z [Filesystem] [info] Shutting down server... { metadata: undefined }
2025-07-23T13:16:33.049Z [Filesystem] [info] Initializing server... { metadata: undefined }
2025-07-23T13:16:33.092Z [Filesystem] [info] Using built-in Node.js for MCP server: Filesystem { metadata: undefined }
2025-07-23T13:16:33.462Z [Filesystem] [info] Server started and connected successfully { metadata: undefined }
2025-07-23T13:16:33.737Z [Filesystem] [info] Message from client: {"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"claude-ai","version":"0.1.0"}},"jsonrpc":"2.0","id":0} { metadata: undefined }
2025-07-23T13:16:33.740Z [Filesystem] [info] Server transport closed { metadata: undefined }
2025-07-23T13:16:33.741Z [Filesystem] [info] Client transport closed { metadata: undefined }
2025-07-23T13:16:33.741Z [Filesystem] [info] Server transport closed unexpectedly, this is likely due to the process exiting early. If you are developing this MCP server you can add output to stderr (i.e. console.error('...') in JavaScript, print('...', file=sys.stderr) in python) and it will appear in this log. { metadata: undefined }
2025-07-23T13:16:33.742Z [Filesystem] [error] Server disconnected. For troubleshooting guidance, please visit our debugging documentation { metadata: { context: 'connection', stack: undefined } }
2025-07-23T13:16:33.743Z [Filesystem] [info] Client transport closed { metadata: undefined }
2.
When I set the filesystem to MCP, the initial connection works and fetches the folder from the initial connection point. But when I execute the next command, I get an error like the one in the attached screenshot. I tested this on over five different computers with two different IDs and encountered the same issue. I am eagerly awaiting a prompt update from Claude.
<img width="640" height="757" alt="Image" src="https://github.com/user-attachments/assets/dc5d32ab-c8b9-4f48-807b-3e8e91d05b77" />
<img width="824" height="731" alt="Image" src="https://github.com/user-attachments/assets/2ba7972e-8986-412d-b0ff-5023c4c7ada6" />
Same here :(
+1
Sorry, we can't help with Claude Desktop issues.
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.