BUG: MCP servers fail to connect in Claude Code despite correct configuration
Bug Description
Operating System: macOS Darwin 24.5.0
Node.js Version: v23.10.0 (via nvm)
Issue Description:
MCP servers consistently fail to connect in Claude Code, even though they are properly configured and the servers themselves work correctly when tested independently.
Steps to Reproduce:
- Install MCP server globally:
npm install -g @modelcontextprotocol/server-filesystem
- Add MCP server to Claude Code:
claude mcp add filesystem "npx" "@modelcontextprotocol/server-filesystem" "/Users/username/Projects/"
- Verify server is listed:
claude mcp list
# Output shows: filesystem: npx @modelcontextprotocol/server-filesystem /Users/username/Projects/
- Start Claude Code:
claude
- Attempt to use MCP tools - all fail with "MCP servers failed to connect"
Expected Behavior:
MCP servers should connect and be available for use within Claude Code.
Actual Behavior:
All MCP servers fail to connect, despite being properly configured and working when tested independently.
Additional Information:
- The MCP server works correctly when tested directly:
echo '{"jsonrpc":"2.0","method":"initialize","params":{"protocolVersion":"2024-11-05","clientInfo":{"name":"test","version":"1.0.0"},"capabilities":{}},"id":1}' | npx @modelcontextprotocol/server-filesystem /path/to/directory
- Returns successful initialization response.
- Tried multiple configuration approaches:
- Using npx with -y flag
- Using direct node path with full path to server
- Creating .mcp.json in project directory
- Using --mcp-config flag with JSON file
- Importing from Claude Desktop config
- No error logs are generated in any standard log locations.
- The same configuration works in Claude Desktop app but fails in Claude Code CLI.
Environment Details:
- Multiple MCP servers tested (filesystem, memory, github, etc.)
- All exhibit the same connection failure
- No permission issues (all files have correct permissions)
- PATH includes node/npm directories
This appears to be a fundamental issue with Claude Code's MCP integration rather than a configuration problem, as the servers work correctly outside of Claude Code.
Environment Info
- Platform: macos
- Terminal: pycharm
- Version: 1.0.11
- Feedback ID: 74d37a8d-973a-43c3-96ea-e43aa60b7888
Errors
[{"error":"AbortError: The operation was aborted\n at abortChildProcess (node:child_process:716:27)\n at onAbortListener (node:child_process:786:7)\n at process.processTicksAndRejections (node:internal/process/task_queues:85:11)","timestamp":"2025-06-04T23:03:51.295Z"},{"error":"Error: Command failed: security find-generic-password -a $USER -w -s \"Claude Code\"\nsecurity: SecKeychainSearchCopyNext: The specified item could not be found in the keychain.\n\n at genericNodeError (node:internal/errors:983:15)\n at wrappedFn (node:internal/errors:537:14)\n at checkExecSyncError (node:child_process:882:11)\n at execSync (node:child_process:954:15)\n at JZ (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:683:2402)\n at file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:582:11763\n at Q (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:526:17069)\n at LU (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:582:10909)\n at HG (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:582:10355)\n at lO2 (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:2042:4758)","timestamp":"2025-06-04T23:03:51.321Z"},{"error":"Error: No logs found at /Users/davidandrews/Library/Caches/claude-cli-nodejs/-Users-davidandrews-PycharmProjects-lexgenius/messages\n at tT4 (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:583:3738)\n at iQ1 (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:583:3487)\n at OAA.<anonymous> (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:2337:2015)\n at process.processTicksAndRejections (node:internal/process/task_queues:105:5)\n at async OAA.parseAsync (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:517:4184)\n at async vE5 (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:2340:5983)\n at async xE5 (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:2336:17530)","timestamp":"2025-06-04T23:03:51.349Z"},{"error":"AbortError: The operation was aborted\n at abortChildProcess (node:child_process:716:27)\n at AbortSignal.onAbortListener (node:child_process:786:7)\n at [nodejs.internal.kHybridDispatch] (node:internal/event_target:827:20)\n at AbortSignal.dispatchEvent (node:internal/event_target:762:26)\n at runAbort (node:internal/abort_controller:449:10)\n at abortSignal (node:internal/abort_controller:435:3)\n at AbortController.abort (node:internal/abort_controller:468:5)\n at Timeout._onTimeout (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:2336:15943)\n at listOnTimeout (node:internal/timers:611:17)\n at process.processTimers (node:internal/timers:546:7)","timestamp":"2025-06-04T23:03:54.232Z"},{"error":"ShellError: Shell command failed\n at Object.call (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1547:1326)\n at process.processTicksAndRejections (node:internal/process/task_queues:105:5)\n at async pZ5 (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1889:19078)\n at async YK1 (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1889:17664)\n at async dZ5 (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1889:16706)\n at async hZ5 (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1889:16356)\n at async oe (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1889:15392)\n at async oe (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1889:16218)\n at async oe (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1889:16218)\n at async oe (file:///Users/davidandrews/.nvm/versions/node/v23.10.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1889:16218)","timestamp":"2025-06-04T23:05:14.363Z"},{"error":"ShellError: Shell command failed\n at Object.call (file:///Users/davidandrews/.nvm/versions/nod
Note: Error logs were truncated.
48 Comments
was this fixed? i'm still having issues with claude code connecting to supabase mcp
Same issue for basically any mcp server
Experiencing the same issue.
Restarting Claude several times and eventually it connects. But it's almost like the MCP server needs to be 'warmed up'?
A restart of claude, initiazes the app but page is blank. Go to settings, then closing settings, and then the page populates.
macOS: Sequoia 15.5
node version v22.16.0
UPDATE: Ok, folks. So some complexity here.
npm --versionwould returned a 'not recognized'.I'm still testing, but so far seems like it's working much faster and bettah.
In short;
I have the same issue .I just had the demo server code in demo server python sdk . It also give me cant connect to the mcp server. Any suggestions?
Can you try running
claude --debugto see if anything is displayed?i tried to but i cant get to run claude debug .cli is not found and I dont think its the issue since im using python sdk.
Having the same problem with a couple of MCP servers.
wasted hours trying to get gitea MCP working with claude-code
I'm also seeing this. The MCP server works fine in Claude Desktop but Claude Code can't find any tools for it
Having the same issue - It will work fine for a few tool uses, and then spontaneously disconnect and can't reconnect with restarting claude code, sometimes a few times. MacOS.
This is still an issue and a very breaking issue. It should be prioritized
Hey all, I'm trying to understand all of the reports here. Is this a single issue or are there multiple things going on here?
@timjrobinson can you share more about the server config you're using? Does the server show as connected in
/mcpbut says it has no tools when you drill into the details?@seanmozeik are you using stdio servers or remote servers? Or is this for any server?
If anyone is able to run with
claude --debugand share output from that, that'll really help. You can also find MCP logs in~/Library/Caches/claude-cli-nodejs/(within different subdirectories there).Mine just timeout after the limit is reached . The Servers work fine on Claude Desktop. I import them over to claude code using claude mcp add-from-claude-desktop command.
Sometimes the servers work in Claude Code, other times more often than not they just dont work. Very buggy
[DEBUG] MCP server "mcp-server-firecrawl": Connection failed: Error: Connection to MCP server "mcp-server-firecrawl" timed out after 30000ms
[DEBUG] MCP server "mcp-server-firecrawl": Error message: Connection to MCP server "mcp-server-firecrawl" timed out after 30000ms
[DEBUG] MCP server "mcp-server-firecrawl": Error stack: Error: Connection to MCP server "mcp-server-firecrawl" timed out after 30000ms
at Timeout._onTimeout (file:///home/username/.nvm/versions/node/v22.16.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1383:5751)
at listOnTimeout (node:internal/timers:588:17)
at process.processTimers (node:internal/timers:523:7)
╭────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
[DEBUG] MCP server "firebase-mcp": Connection failed: Error: Connection to MCP server "firebase-mcp" timed out after 30000ms
[DEBUG] MCP server "firebase-mcp": Error message: Connection to MCP server "firebase-mcp" timed out after 30000ms
[DEBUG] MCP server "firebase-mcp": Error stack: Error: Connection to MCP server "firebase-mcp" timed out after 30000ms
at Timeout._onTimeout (file:///home/username/.nvm/versions/node/v22.16.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1383:5751)
at listOnTimeout (node:internal/timers:588:17)
at process.processTimers (node:internal/timers:523:7)
╭────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
[DEBUG] MCP server "playwright": Connection failed: Error: Connection to MCP server "playwright" timed out after 30000ms
[DEBUG] MCP server "playwright": Error message: Connection to MCP server "playwright" timed out after 30000ms
[DEBUG] MCP server "playwright": Error stack: Error: Connection to MCP server "playwright" timed out after 30000ms
at Timeout._onTimeout (file:///home/username/.nvm/versions/node/v22.16.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1383:5751)
at listOnTimeout (node:internal/timers:588:17)
at process.processTimers (node:internal/timers:523:7)
╭────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
[DEBUG] MCP server "youtube-transcript": Connection failed: Error: Connection to MCP server "youtube-transcript" timed out after 30000ms
[DEBUG] MCP server "youtube-transcript": Error message: Connection to MCP server "youtube-transcript" timed out after 30000ms
[DEBUG] MCP server "youtube-transcript": Error stack: Error: Connection to MCP server "youtube-transcript" timed out after 30000ms
at Timeout._onTimeout (file:///home/username/.nvm/versions/node/v22.16.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1383:5751)
at listOnTimeout (node:internal/timers:588:17)
at process.processTimers (node:internal/timers:523:7)
╭────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
[DEBUG] MCP server "ElevenLabs": Connection failed: Error: Connection to MCP server "ElevenLabs" timed out after 30000ms
[DEBUG] MCP server "ElevenLabs": Error message: Connection to MCP server "ElevenLabs" timed out after 30000ms
[DEBUG] MCP server "ElevenLabs": Error stack: Error: Connection to MCP server "ElevenLabs" timed out after 30000ms
at Timeout._onTimeout (file:///home/username/.nvm/versions/node/v22.16.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1383:5751)
at listOnTimeout (node:internal/timers:588:17)
at process.processTimers (node:internal/timers:523:7)
╭────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮[DEBUG] MCP server "qdrant": Connection failed: Error: Connection to MCP server "qdrant" timed out after 30000ms
[DEBUG] MCP server "qdrant": Error message: Connection to MCP server "qdrant" timed out after 30000ms
[DEBUG] MCP server "qdrant": Error stack: Error: Connection to MCP server "qdrant" timed out after 30000ms
at Timeout._onTimeout (file:///home/username/.nvm/versions/node/v22.16.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1383:5751)
at listOnTimeout (node:internal/timers:588:17)
at process.processTimers (node:internal/timers:523:7)
╭────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
[DEBUG] MCP server "postgres": Connection failed: Error: Connection to MCP server "postgres" timed out after 30000ms
[DEBUG] MCP server "postgres": Error message: Connection to MCP server "postgres" timed out after 30000ms
[DEBUG] MCP server "postgres": Error stack: Error: Connection to MCP server "postgres" timed out after 30000ms
at Timeout._onTimeout (file:///home/username/.nvm/versions/node/v22.16.0/lib/node_modules/@anthropic-ai/claude-code/cli.js:1383:5751)
at listOnTimeout (node:internal/timers:588:17)
at process.processTimers (node:internal/timers:523:7)
@ashwin-ant the MCP server has started working for me again after adding it in Claude code explicitly with
claude mcp addit can see the tools. Previously when using the claude file from claude desktop it wasn't working (I guess code and desktop use different configs and I didn't realize?)Can confirm I also have this issue with FastMCP
stdiotransport. I've been trying this for a while, I've wrapped calls and a lot more, but for whatever reason it really struggles.this is Python FastMCP, but it is consistently happening in my package with Claude Code, and then the entire MCP server crashes after, which is the real issue, because then I can't use it. Not sure if it's directly connected, but I would love to contribute to a fix
I'm running claude code on wsl and claude seems to intermitently detect or fail to detect by project scoped mcp's in .mcp.json. This is a pain as i want to keep my MCP config project scoped and stored with the project in github
The filesystem mcp, probably one of the most widely used ones still refuses to connect.. I love Claude Code but this breaking constantly is very annoying.
did update to. v1.0.44, and all my 11 mcps stopped working instantly, even the custom ones, also claude code started crashing on "escape" but i can see there are other issues mentioning this
I have the same issue, running in WSL with MCPs for Supabase and Context7 on a NextJS project. Gemini CLI works fine and Claude code works fine in other projects.
Im sorry but this MCP thing needs to be sorted out. The MCP concept is great but I've rarely seen a tech so poorly handled, this is on par with javascript libraries in the 10s. MCP is a mess. It is unreliable, every vendor has a different implementation. Its frankly too annoying, and Anthropic has the least excuses to be getting this wrong still.
Not only do they make it ridiculously annoying on Claude Code and Claude Desktop but its not rock solid. CLine and Roo have MCP marketplace, one click and install. ANd its mostly working fine. I need CC and CD, but I also need reliable and easy MCP.
For the love of god, stop making new features and make the basics work reliably please.
I hope Claude isn’t writing the MCP code 😢
Shoddy work. Very very expensive.
No motivation to cache parses which burn up all my tokens.
Also experiencing this issue with MCPs not available in Claude code. Any updates on a fix here?
I'm on v1.0.44 - I've just moved the scope of my MCP's from project to user and that has worked - not ideal, but at least they work
We're investigating a potential fix. Some questions:
npxcommands or as localnode/dockercommands? Or a mix? It sounds like it's both, but want to make sure.Multiple MCP servers some at users level, some at project level in .mcp.json configured with NPX commands. @AndyDavo suggestion worked for me
I just installed Claude Code today; also getting the mcp failure to connect. I had these two:
It's still not working if I remove either one and try with just one.
I have the following;
Mostly it's working fine.
I have a;
I also have all the bells and whistled turned on; IE
Here is the error from the mcp-server log for one of my obsidian mcp's. I'm guessing a caching problem but not sure.
I'm also getting a message to upgrade npm but at this point I'm kinda afraid to.
Current version: 10.9.2
2025-07-10T14:58:42.412Z [mcp-obsidian-linnaean] [info] Server started and connected successfully { metadata: undefined }
2025-07-10T14:58:42.416Z [mcp-obsidian-linnaean] [info] Message from client: {"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"claude-ai","version":"0.1.0"}},"jsonrpc":"2.0","id":0} { metadata: undefined }
npm error code ENOTEMPTY
npm error syscall rename
npm error path /Users/userName/.npm/_npx/0496d3f111c50e47/node_modules/@modelcontextprotocol/sdk
npm error dest /Users/userName/.npm/_npx/0496d3f111c50e47/node_modules/@modelcontextprotocol/.sdk-SXsa55pC
npm error errno -66
npm error ENOTEMPTY: directory not empty, rename '/Users/userName/.npm/_npx/0496d3f111c50e47/node_modules/@modelcontextprotocol/sdk' -> '/Users/userName/.npm/_npx/0496d3f111c50e47/node_modules/@modelcontextprotocol/.sdk-SXsa55pC'
npm error A complete log of this run can be found in: /Users/userName/.npm/_logs/2025-07-10T14_58_43_527Z-debug-0.log
Had to reinstall all MCPs after they crashed following a recent update. My ~/.claude.json also got corrupted (ballooned to 3MB+ from paste history).
Here's my current MCP setup after rebuilding everything:
🌍 Global User Level (~/Library/Application Support/Claude/claude_desktop_config.json):
📁 Project Level (.mcp.json):
💻 Local/Session Level (.claude-code-mcp.json + claude mcp list):
📝 Additional Local Config (.claude-code-mcp.json):
Note: devlog is my custom MCP with ChromaDB vector search included. Some MCPs are running through Docker containers.
When I use the same .mcp.json file in ubuntu claude code works fine. However, in WSL under Windows, Ubuntu prompts that it cannot be connected
Just my local gitea server, which Gemini connected to immediately.
https://gitea.com/gitea/gitea-mcp
just
sudo chown -R 501:20 "$HOME/.npm"
or
Try executing the command in the mcp server settings and check the log.
ex) zsh) npx -y ios-simulator-mcp
Heads up to other devs using MCP servers with Claude Code:
Try running
claude --verboseonce and watch what happens at startup. You might be surprised.I started investigating this after I almost hit all my token limits surprisingly quickly in a previous session. I discovered that ALL those MCP tool descriptions are being loaded into your context immediately, whether you use verbose mode or not. The verbose flag just makes it visible.
This means if you've been wondering why you're hitting token limits faster lately, this is probably why. Every session starts with potentially thousands of tokens already consumed by tool descriptions you might never use.
It's worth checking out with
--verbosejust to see the actual impact. Pretty eye-opening to see how much context is being eaten up before you even start coding.Based on my research, MCP tool descriptions used to load on-demand (only when you actually use the tool), which seems like the intended behavior. This appears to be an unintended change in recent versions.
I'm currently optimizing my MCP setup to only keep the most essential tools enabled until this gets sorted out. The Claude Code team has built an amazing tool, and I'm sure they'll address this once they're aware of the impact.
Hope this helps others understand their token usage better!
I also have this issue. MCP servers work in VS Code and Cursor IDE, but Claude Code tells me it's not exposed to the tools and that it's a CC API issue.
After some tests, I see that claude totally ignores any MCP servers defined in the local project's .claude.json and .mcp.json. That is very easy to check, just create or edit that file, add some valid MCP server definition, and run "claude mcp list", it will not list whatever you added there.
Instead, it only reads ~/.claude.json, which has a list of MCP servers per each path it was used and each parent directory until the user's home directory.
Same issue, after hours trying, i’ve been UNABLE TO USE NOTION MCP from Claude CODE, despite that it’s 55% of the times functional on browser and claude DESKTOP.
Claude Code MCP is broken across ALL platforms?
Skip for the moment from using MCP in Claude Code entirely, regardless of OS?
2 days i cant setup any mcps inside claude code. The command in terminal "claude mcp list" shows that i have working mcps,but the command in CC "/mcp" shows 'No MCP servers configured'. I did everything and it always shows 'No MCP servers configured' error.
OS: Windows 10
Installation: npm-global (1.0.62)
Problem: "The interactive session of Claude Code does not load any user-configured MCP servers (neither from a project-level .mcp.json file nor from the global configuration set by claude mcp commands). The /mcp command shows 'No MCP servers configured', while the external claude mcp list command correctly shows connected servers. Debug logs confirm that only the built-in 'ide' MCP server is connected, with no attempts to load any others."
I think it's all in beta, not worth to waste time yet on this.
Starting Claude with --mcp-config /home/user/project-dir/.claude/settings.json shows the mcp server in Ubuntu 24.04. Without the said flag and path specification, the project MCP server doesn't get found.
I had the same issue and saw this lines on .claude\settings.local.json on my local folder, deleting them made CC aware of the MCPs
"disabledMcpjsonServers": [
"supabase",
"chrome-devtools"
]
Thanks, I'll try that asap!
Il giorno sab 2 ago 2025 alle ore 12:28 spidey000 @.***>
ha scritto:
sad to see claude is that buggy. wasted hours with no solution.
I'm having a similar but not quite identical issue, dropping a comment here in case it helps:
I've built an MCP server for my specific project, and set it up in .mcp.json. Claude Code sees the server, but it sees no capabilities, and as a result doesn't show any tools. I can't really install this server globally to tell if the local/global distinction makes a difference because it's very specific to my project and in fact actually detects when it's running from the target/ directory to enable debug mode.
The server seems to run fine. Claude wrote most of the server itself. If I manually send the capability/tool listing requests from the spec, things look fine. Claude even wrote a bunch of test shell and Python scripts to test the MCP server and claims the bug is on your end. :)
So this isn't quite the server not working--it shows as connected, but with no capabilities and thus no tools.
I'll note that initially the capabilities were empty until I discovered that they needed to contain "tools". I don't know if there's any caching happening--maybe my server is still referenced as exposing no tools even though I've fixed that bug on my end--but I've searched all the JSON config files I could find in my project/~/.claude* and didn't find anything obvious.
I also have the Puppeteer MCP server installed globally and it works fine. Unfortunately again, I can't install this one globally because it's specific to my project.
Make sure you kill all prior Claude tasks before restarting the MCP server. This had me stumped for a few hours yesterday. This Youtube video was helpful - https://www.youtube.com/watch?v=xcj_5xl1RNs&t=36s
Got it working finally. Lots of steps, including needing to run the outlook-auth-server.js locally via node.exe and open port 3333 to the internal network on the Windows firewall. I used Claude to help me troubleshoot and it saved a whole lot of time.
Here is the MCP project I used to connect Outlook / M365 to Claude - https://github.com/ryaker/outlook-mcp
Disclaimer - I have been in tech since 1983 but never became a coder. Currently in working in cybersecurity.
claude --debughelped me figure it out that i was missing docker
I'm having the same issue.
<img width="1350" height="334" alt="Image" src="https://github.com/user-attachments/assets/6cf14462-5ce7-4a91-8a48-2d9031cbbbc2" />
<img width="836" height="352" alt="Image" src="https://github.com/user-attachments/assets/186c0f36-48d1-42a2-bc1b-9a3c5ff8c7d8" />
I keep getting: Error: Not connected
And no. the issue is NOT my backend it is working via swagger etc.
<img width="1158" height="1116" alt="Image" src="https://github.com/user-attachments/assets/8dec660c-b7c5-4269-8173-72d2d83517d4" />
As you can see my swagger is working perfectly so idk what the issue is.
EDIT: Of course it works when I start claude via --debug. Of course!!!!!
For folks hitting similar issues, debug by:
/mcpto verify whether your MCP server is found, and what its current status isclaude --debug) for more detailed MCP error output during startupClosing this issue. Please feel free to open a new issue if the steps above don't help.
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.