Command-Based MCP Servers Not Exposing Tools to AI Assistant

[Lee_Sargent]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Command-Based MCP Servers Not Exposing Tools to AI Assistant

SUMMARY
Command-based MCP servers (using command + args) successfully connect and discover tools, but those tools are not exposed to the AI assistant or displayed in the UI. URL-based MCP servers work correctly.

ENVIRONMENT

• Cursor Version: [Your version - check Help > About]
• OS: macOS 24.6.0 (darwin)
• MCP Server Type: Command-based (npx with mcp-remote)

CONFIGURATION
{
“mcpServers”: {
“atlassian”: {
“command”: “npx”,
“args”: [“-y”, “[email protected]”, “https://mcp.atlassian.com/v1/sse”]
},
“Sentry”: {
“url”: “https://mcp.sentry.dev/mcp”
}
}
}

STEPS TO REPRODUCE

1. Configure a command-based MCP server (e.g., Atlassian) in ~/.cursor/mcp.json
2. Restart Cursor
3. Open MCP settings panel
4. Observe that the server shows “No tools, prompts, or resources” despite being enabled
5. Check logs - tools are discovered but not exposed

EXPECTED BEHAVIOR

• Tools should be displayed in the MCP settings UI
• Tools should be accessible to the AI assistant
• list_mcp_resources() should return available tools

ACTUAL BEHAVIOR

• UI shows “No tools, prompts, or resources”
• Tools are discovered (logs show “Found 27 tools”)
• Tools are cached in mcp-cache.json
• Tools are NOT accessible to AI assistant
• list_mcp_resources() returns empty

EVIDENCE

Logs Show Tools Are Discovered
From MCP user-atlassian.log:

2025-12-10 15:41:40.833 [info] listOfferings: Found 27 tools
2025-12-10 15:42:26.710 [info] listOfferings: Found 27 tools

Tools Are Cached
~/.cursor/projects/[project]/mcp-cache.json contains:

{
“atlassian”: {
“tools”: [
{
“name”: “atlassianUserInfo”,
“description”: “Get current user info from Atlassian”,
…
},
// … 26 more tools
]
}
}

Connection Is Working
Logs show:

• “Proxy established successfully”
• “Connected to remote server using SSEClientTransport”
• “Successfully connected to stdio server”
• Authentication tokens are valid

Comparison: URL-Based Server Works
Sentry MCP (URL-based) works correctly:

• Tools are displayed in UI
• Tools are accessible to AI assistant
• list_mcp_resources() works

ROOT CAUSE HYPOTHESIS
Command-based MCP servers are successfully:

1. Connecting to the server
2. Discovering tools
3. Caching tool definitions

But failing to:

1. Register tools with the AI assistant layer
2. Display tools in the UI
3. Expose tools via MCP resource APIs

This suggests a bug in how Cursor handles command-based MCP servers vs URL-based servers.

WORKAROUND
None currently. Must use:

• Direct REST API calls
• Jira web interface exports
• Other non-MCP integration methods

ADDITIONAL INFORMATION

• MCP server: Atlassian Rovo MCP Server
• mcp-remote version: 0.1.13
• Authentication: OAuth flow completes successfully
• Multiple tools discovered: 27 tools (Jira + Confluence)

REQUESTED FIX
Command-based MCP servers should expose discovered tools to:

1. The AI assistant (make tools callable)
2. The UI (display tools in MCP settings)
3. MCP resource APIs (return tools via list_mcp_resources())

Steps to Reproduce

STEPS TO REPRODUCE

1. Configure a command-based MCP server (e.g., Atlassian) in ~/.cursor/mcp.json
2. Restart Cursor
3. Open MCP settings panel
4. Observe that the server shows “No tools, prompts, or resources” despite being enabled
5. Check logs - tools are discovered but not exposed

EXPECTED BEHAVIOR

• Tools should be displayed in the MCP settings UI
• Tools should be accessible to the AI assistant
• list_mcp_resources() should return available tools

ACTUAL BEHAVIOR

• UI shows “No tools, prompts, or resources”
• Tools are discovered (logs show “Found 27 tools”)
• Tools are cached in mcp-cache.json
• Tools are NOT accessible to AI assistant
• list_mcp_resources() returns empty

EVIDENCE

Logs Show Tools Are Discovered
From MCP user-atlassian.log:

2025-12-10 15:41:40.833 [info] listOfferings: Found 27 tools
2025-12-10 15:42:26.710 [info] listOfferings: Found 27 tools

Tools Are Cached
~/.cursor/projects/[project]/mcp-cache.json contains:

{
“atlassian”: {
“tools”: [
{
“name”: “atlassianUserInfo”,
“description”: “Get current user info from Atlassian”,
…
},
// … 26 more tools
]
}
}

Connection Is Working
Logs show:

• “Proxy established successfully”
• “Connected to remote server using SSEClientTransport”
• “Successfully connected to stdio server”
• Authentication tokens are valid

Comparison: URL-Based Server Works
Sentry MCP (URL-based) works correctly:

• Tools are displayed in UI
• Tools are accessible to AI assistant
• list_mcp_resources() works

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.2.9 (Universal)
VSCode Version: 1.105.1
Commit: a86689c93e9fb11addfbefd29a6ec7c0a59175e0
Date: 2025-12-10T16:51:25.713Z (3 hrs ago)
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Darwin arm64 24.6.0

Does this stop you from using Cursor

No - Cursor works, but with this issue

[emulix (Eldar Mulic)]

Hitting the same issue on Windows as well on the same version. This mcp was buggy before this version but at least working. Now it’s not working at all.

Version: 2.2.9 (user setup)
VSCode Version: 1.105.1
Commit: a86689c93e9fb11addfbefd29a6ec7c0a59175e0
Date: 2025-12-10T16:51:25.713Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.26200

[deanrie]

Hey, thanks for the report. This is a known bug in version 2.2.9.

Please update to version 2.2.14 (Help > Check for Updates). It may include a fix for command-based MCP servers.

After updating, restart Cursor and check if the tools appear in the UI and become available to the AI. Let me know if this resolves the issue.