Desktop app MCP Tools hanging for 5-10 minutes then failing.
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?
Environment
- Product: Claude Desktop (Windows GUI app, NOT Claude Code CLI)
- Version: 1.1.1520 (48ced7)
- OS: Windows 11
- Date Started: February 1, 2026
Bug Description
ALL MCP tool calls hang for exactly ~5 minutes, then fail with "No result received from client-side tool execution". This makes Claude Desktop completely unusable for any MCP-based workflows.
Reproduction Steps
- Configure any MCP server (SSH, filesystem, etc.)
- Attempt to use MCP tool
- Observe 5+ minute hang
- Tool fails with "No result received from client-side tool execution"
Tested MCP Servers (ALL FAIL):
- SSH server (custom tradasite server to 91.99.146.87)
- Filesystem servers
- All operations timeout
Timing Evidence
Test performed: 21:20:05 → 21:25:46 = 5 minutes 41 seconds hang
Expected Behavior
MCP tools should execute within seconds, not hang for 5+ minutes
Actual Behavior
- 5+ minute freeze
- No progress indication
- Eventually fails with "No result received"
- Operation never executes on server
Screenshots
[Attach your 3 screenshots showing the hang and timestamps]
Related Issues
May be related to Claude Code MCP hanging bugs reported Jan 31, 2026:
- #22158 (v2.1.27 hangs at 100% CPU)
- #22172 (v2.1.23+ parallel instance hangs)
- #15945 (CRITICAL MCP Server 16+ hour hang)
Impact
CRITICAL - Production blocking
- Unusable for development workflows
- All MCP functionality broken
- No workaround available
Request
Emergency hotfix needed - this version is completely broken for MCP users.
What Should Happen?
Claude should connect to my server instantly (5 seconds), perform coding tasks, and summerise without delays.
Error Messages/Logs
The command didn't return output. Let me try a simpler search.
No result received from client-side tool execution.
Steps to Reproduce
Steps to Reproduce
- Install Claude Desktop version 1.1.1520 (Windows)
- Configure a simple filesystem MCP server in
%APPDATA%\Claude\claude_desktop_config.json:
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"C:\\Users\\%USERNAME%\\Documents"
]
}
}
}
- Restart Claude Desktop completely (close and reopen)
- Verify MCP server connected - check for hammer icon (🔨) in chat input box
- Start timer - note current time (HH:MM:SS)
- Send this prompt:
Please use the filesystem MCP tool to list all files in Documents folder
- Observe the hang:
- MCP tool call is initiated
- Application shows spinner/loading state
- Application becomes unresponsive
- Wait and time it - will hang for approximately 5+ minutes
- Eventually fails with error: "No result received from client-side tool execution"
- Record end time - note when error appears (HH:MM:SS)
- Calculate duration - should be approximately 5-6 minutes
Minimal Example
Config file location: %APPDATA%\Claude\claude_desktop_config.json
Config content:
{
"mcpServers": {
"test-filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "C:\\temp"]
}
}
}
Test prompt: "Use the test-filesystem MCP to create test.txt with content 'hello'"
Expected: File created in ~1-5 seconds
Actual: 5+ minute hang, then failure with "No result received"
Timing Evidence from Real Test
- Start: 21:20:05
- End: 21:25:46
- Duration: 5 minutes 41 seconds
- Result: Operation failed, MCP tool never executed
Reproduced On
- Windows 11
- Claude Desktop v1.1.1520 (48ced7)
- Multiple different MCP servers (filesystem, SSH, custom servers)
- All MCP tools exhibit identical ~5 minute timeout behavior
Claude Model
Sonnet (default)
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
1.1.1520 (48ced7)
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
PowerShell
Additional Information
_No response_
20 Comments
Same as
https://github.com/anthropics/claude-code/issues/22299
and
https://github.com/anthropics/claude-code/issues/22319
I looked further into it and the MCP server works once per tool call the next call never makes it to the server. Any MCP server. You can see other threads on the Claude Chrome extension basically failing the same way. All Claude Desktop and Chrome MCP servers are affected.
my desktop app hasn't been able to read my filesystem for almost 3 days now.
Same issue here (MacOS), filesystem access is completely broken
The workaround from Reddit is to actually open the collapesed tool calls and they will go through:
https://www.reddit.com/r/ClaudeAI/comments/1qtspwj/claude_will_not_think_until_i_click_the_collapse/
Why the MCP is tied to the UI like that the world will never know.
Ok , MCP tools are not working properly. I use Desktop Commander and now it works sporadically and hangs a lot. I wish Claude Desktop commander would ask me if I want to install an new version . Maybe it did :-) But now I need to revert to an older version - kind of annoying!
Resumed coding after a week and was frustrated to see this issue. The tool call for locating, reading, moving or making changes to the files all have become extremely slow. They dont progress unless you expand the tool call in the chat. I am using Claude Chat on Claude desktop and heavily rely on the MCP extension to make changes and directly work w the codebase.
Has this issue propped up recently (past couple of days)? Has anyone had any luck from reinstalling the connectors or any other methods?
Yes, I've tried reinstalling everything. Every MCP connection and Claude Desktop and the issue still persists. This has been an issue since Friday. I read on another bug report that rolling back Claude Desktop versions doesn't help either. 4 days without MCP tools is inexcusable on a $200/mo subscription.
macOS confirmation + server-side evidence that calls never leave the client
Environment: macOS Sequoia (Darwin 24.6.0), 4 custom Python MCP servers (stdio transport), Claude Desktop latest
Server-side proof: tool calls never reach the servers
I have Langfuse tracing wired into all my MCP server
call_tool()handlers. During failures, I queried the Langfuse REST API to check whether tool calls actually reached the server processes. Key finding:ps aux)This confirms the bug is entirely client-side. The servers are healthy and waiting; Claude Desktop just isn't sending the messages down the stdio pipe.
Controlled workaround test: expanding tool call UI boxes
Confirming @sonofsmog's workaround from the Reddit thread — clicking to expand the MCP tool call UI boxes unblocks the stdio pipe. Tested systematically:
| Round | Method | Boxes expanded? | Result |
|-------|--------|----------------|--------|
| 1 | Same 4 tools across 4 servers | No | 3 of 4 failed (only Morgan succeeded). Langfuse: 1 trace. |
| 2 | Same 4 tools | Yes (clicked each) | 4/4 succeeded |
| 3 | Same 4 tools | Yes | 4/4 succeeded |
| 4 | Same 4 tools | Yes | 4/4 succeeded |
| 5 | 4 completely different methods | Yes | 4/4 succeeded on first call |
Without expanding: failures. With expanding: 100% success across 5 consecutive rounds including novel methods. The workaround is reliable on macOS.
Visible symptom: slow text rendering correlates with blocked pipe
When tool calls are about to fail, Claude Desktop's text output slows to ~1 character per 5 seconds for the first few words, then suddenly "snaps" back to normal speed before displaying the failure. This happened consistently across multiple failure instances and suggests the UI render thread is blocked on the stale stdio read.
Selective failure mode (not reported by others yet)
In addition to the all-tools-fail scenario, I observed a selective failure mode: Claude Desktop called 4 different MCP servers, but only 1 responded (Morgan). The other 3 failed silently. Langfuse confirmed only the successful call generated a trace. This suggests each server's stdio connection is managed independently and can fail independently — it's not a global lock.
I noticed that but it still fails over and over again trying to read the filesystem and eventually crashes and deletes the chat of what it was trying to do.
I'm the author of that Reddit thread. I haven't had a single failure on manually expanding the Tool tree to push the call through. What errors/behaviour are you seeing?
<img width="1101" height="1070" alt="Image" src="https://github.com/user-attachments/assets/ca3393fe-5c45-40e8-89d2-5d6ec356ec9a" />
Thank you everyone for the report! We turned off a feature flag that was potentially causing the issue. Can you try again and see if it's working?
Really apologize for the disruption and appreciate you guys for taking the time to issue!
✅ Confirmed FIXED for Claude Desktop app — macOS
Following the feature flag fix mentioned by @alicelovescake, I can confirm the issue is resolved for Claude Desktop app on macOS.
Test Environment
Test Results (this conversation)
| # | Provider | Tool tested | Result |
|---|----------|-------------|--------|
| 1 | Desktop Commander |
list_directory| ✅ instant || 2 | Control your Mac |
osascript| ✅ instant || 3 | Claude in Chrome |
tabs_context_mcp| ✅ instant || 4 | Linux Container |
bash_tool| ✅ instant || 5 | Desktop Commander |
get_file_info| ✅ instant || 6 | Control your Mac |
osascript(2nd call) | ✅ instant |Conversation Statistics — Tool Calls Audit
| Category | Count |
|----------|-------|
| conversation_search | 2 |
| Desktop Commander calls | ~15 |
| Control your Mac (osascript) | ~8 |
| Claude in Chrome calls | ~12 |
| bash_tool (container) | ~4 |
| web_search | 2 |
| Total tool calls | ~50+ |
| User blocking clicks | 0 |
| Human wait time | 0 min |
Comparison: Before vs After Fix
| Metric | Before fix (2026-02-01) | After fix (2026-02-03) |
|--------|------------------------|------------------------|
| Tool calls | ~60 | ~50+ |
| Productive % | 33% | 100% |
| Human clicks required | 5-8 | 0 |
| Human wait time | ~24 min | 0 min |
| AI moves wasted on blocking | ~38 (~40%) | 0 |
| App restarts needed | 1 | 0 |
Human Time Cost Analysis
| Segment | Duration | Activity |
|---------|----------|----------|
| 05:45-05:50 | 5 min | Initial test + conversation_search context loading |
| 05:50-05:55 | 5 min | Navigate to #22127, scroll, review comments |
| 05:55-06:00 | 5 min | Navigate to #22451, test MCP tools, draft comment |
| Total | ~15 min | 100% productive (vs ~45 min with blocking before) |
Note
This confirms the fix works for Claude Desktop app on macOS. Windows users may want to confirm separately.
Thank you @anthropic team for the quick turnaround on the feature flag fix! 🙏
---
Tested via Monwoo development workflow — Full diagnostic report in #22127
@Anthropic
For MacOS Desktop App:
<img width="106" height="59" alt="Image" src="https://github.com/user-attachments/assets/6d27b68e-669a-4f69-8635-c1b9ef61e8bb" />
<img width="153" height="224" alt="Image" src="https://github.com/user-attachments/assets/cae19bf0-568a-4316-93c8-5f148aa94250" />
Filesystem tools > e.g. Edit file still gets stuck until click and expand.
Affected Tools:
read_file
read_text_file
read_multiple_files
write_file
edit_file
create_directory
list_directory
list_directory_with_sizes
directory_tree
move_file
search_files
get_file_info
list_allowed_directories
Copy file to Claude
Claude Desktop for Windows MCP Server has issues, often caused by stdout not being captured in sub-processes.
This causes stalls and forces Claude to retry using different approaches (PowerShell, CMD.EXE, redirecting stdout to temp file, etc).
In addition, Windows MCP is quite verbose and clogs the context window.
The solution I found is to give Claude access to the Unix toolset, which is included in GIT for Windows.
This speeds Claude up significantly, uses less context and works better, as Claude 'prefers' Linux.
The implementation is an MCP server, 45 LOC, 2-minute install, you can delete Windows MCP.
git_bash_mcp.py
README.md
The fix initially worked after the above message was posted, but the issue is back this morning on MacOS.
Same symptoms - verbose granular output, fails simple read/right/modify/move tools unless expanded, slow implementation, poor user experience etc.
Can this issue/bug be reopened?
Another possibly related consequence is that in the middle of this, if it does compaction, it restarts its entire thread as if it forgot everything.
@pratik-pednekar what MCP are you using? I just tested with FIlesystem MCP in the chat tab and seemed to work without expanding. can you give me any details on repro steps?
Hi Alicelovescake,
There are (for some reason) two filesystems (Filesystem (desktop) and filesystem (local dev)).
<img width="1146" height="723" alt="Image" src="https://github.com/user-attachments/assets/a299e9f6-0291-408e-bbe7-8a7f6efa758a" />
COnfiguring the local dev version looks like this (i believe I might have 2 MCPs, one claude and one added extension). Is this causing the issue. Any guidance on which is more suitable for working with local files?
<img width="1140" height="662" alt="Image" src="https://github.com/user-attachments/assets/1ba50e74-9e27-4d8e-8636-a24f9a1051a2" />
Thanks for the support.
PS - the issue seems to not be there today, but the file access etc. appear to be slower. There seems to be a max compaction message (possibly unrelated issue)
Pratik
If your Claude Desktop MCP tools hang for 5 minutes and fail with 'No result received', the stdio pipe is being blocked by a collapsed UI element. There's a simple workaround involving the tool call boxes in the chat interface: https://cacheoverflow.dev/blog/KMdWLYK9
Thanks for the report. The Desktop hang was traced and the fix shipped in February. The "expand the tool box to unblock it" workaround a few of you found lines up with the cause. Closing this; please reopen if you still see the 5-10 minute hang on a current Desktop build.