Hello ,
I’m experiencing an issue where MCP (Model Context Protocol) tools configured for SSH-MCP are not appearing in the available tools list for the AI assistant, even though the MCP server is properly configured in .cursor/mcp.json.
Problem Description:
I have configured an SSH-MCP server in my .cursor/mcp.json file to enable remote command execution on a server. Important: My SSH MCP worked perfectly before the latest Cursor update. As a senior DevOps engineer, I was successfully using MCP tools to execute commands on remote servers through the AI assistant. However, after the latest update, it stopped working completely. The configuration appears to be correct (unchanged from when it was working), but when I try to use the AI assistant, the MCP tools (specifically the exec tool from ssh-mcp) are not available in the tool list.
Configuration:
My .cursor/mcp.json contains:
{
"mcpServers": {
"ssh-mcp": {
"command": "npx",
"args": [
"-y",
"ssh-mcp",
"--",
"--host=xx.90.46.xx",
"--port=22",
"--user=dockeruser",
"--password=***",
"--timeout=30000",
"--maxChars=none"
]
}
}
}
Expected Behavior:
According to the documentation and prompts, I should be able to use MCP tools like mcp_kwork-ai-gemini-ssh-mcp_exec or similar to execute commands on the remote server through the AI assistant, without needing to use the local terminal.
Actual Behavior:
- The MCP server appears to be configured correctly
- However, when I query the AI assistant, it reports that MCP tools are not available in its tool list
- Only
list_mcp_resourcesandfetch_mcp_resourceare available, but not the actual MCP tool execution functions - The AI assistant cannot find tools that start with
mcp_*or containssh/execin their names
What I’ve Tried:
- Verified the MCP configuration file syntax
- Checked that the ssh-mcp package should be available via npx
- Reviewed documentation for MCP tool naming conventions
- Attempted to use the tools as described in various prompts
Questions:
- Was there a breaking change in the latest update that affected MCP tool availability?
- Are MCP tools supposed to be automatically available to the AI assistant when configured in
.cursor/mcp.json? (They were before the update) - Is there a specific naming convention or prefix I should expect for MCP tools (e.g.,
mcp_*,mcp_<server-name>_<tool-name>)? Has this changed? - Do I need to restart Cursor or perform any additional setup steps after configuring MCP servers? (This wasn’t required before)
- Are there any logs or debugging methods to verify that MCP servers are properly connected and their tools are being exposed?
- Is this a known regression that’s being tracked or fixed?
Environment:
- OS: Windows 10 (build 26100)
- Cursor version: 2.4.21
- MCP server: ssh-mcp via npx
Additional Context:
This is a regression issue - SSH MCP was working perfectly before the latest Cursor update. As a senior DevOps engineer, I rely heavily on this functionality for remote server management tasks. The ability to execute commands on remote servers via MCP through the AI assistant was a critical part of my workflow, and its absence significantly impacts my productivity. This appears to be a breaking change introduced in the recent update.
Any guidance or assistance would be greatly appreciated!
Thank you!