MCPService: Timeout waiting for EverythingProvider (mcp.createClient) while stdio MCP server stays healthy

Support Bug Reports
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Cursor MCP client creation periodically fails with:

Error creating client: Timeout waiting for EverythingProvider with command 'mcp.createClient'

The underlying MCP server process (observed with user-gitlab / @zereight/mcp-gitlab stdio) remains alive and continues serving tool calls successfully. The failure is in Cursor’s MCP client orchestration layer, not the server exiting.

Steps to Reproduce

  1. Configure multiple MCP servers in ~/.cursor/mcp.json and/or marketplace plugins (GitLab stdio, Grafana Cloud HTTP, Atlassian, Slack, Notion, etc.).
  2. Run a long agent session that issues many parallel GitLab MCP tool calls (e.g. 6+ get_merge_request in one turn).
  3. Leave the session open and watch MCP Logs for 30-60 minutes.

Observed pattern:

  • Every ~5 minutes, multiple parallel createClient calls (often 4 at once for the same server identifier).
  • After ~60 seconds: Timeout waiting for EverythingProvider with command ‘mcp.createClient’.
  • Shared MCP process log on the same PID still shows tool_call_done for get_merge_request (400-2400ms).
  • Same timeout pattern hits other servers (e.g. plugin-grafana-cloud-mcp-grafana-cloud) even when GitLab stdio is fine.

Expected Behavior

Periodic MCP health checks / client refresh should complete without error while the shared MCP process is connected and serving tools.

Operating System

MacOS

Version Information

Cursor IDE (macOS). MCP Logs from session 2026-08-26. GitLab MCP @zereight/mcp-gitlab 2.1.43.

Additional Information

Environment: macOS (darwin 25.6.0). GitLab MCP: @zereight/mcp-gitlab 2.1.43 stdio.

Representative log lines:
[MCPService] createClient: identifier=“user-gitlab”, serverName=“gitlab”
(60s later)
[MCPService] Error creating client: Timeout waiting for EverythingProvider with command ‘mcp.createClient’

Meanwhile [Shared MCP process] PID unchanged:
tool_call_start / tool_call_done: get_merge_request (2000ms+)

Parallel createClient burst example (18:50:08): 4x createClient for user-gitlab, then 4x timeout at 18:51:08.

Impact: Agent loses reliable MCP access mid-session; user must restart MCP or toggle servers. Reads/writes fail in UI while server process is still running.

Possibly related to forum thread on MCP “connection storm” (tools/list cache / reconnect): MCP server regression - does not reload tools, "disconnect" does nothing, ignores mcp.json changes, etc

Workarounds: restart MCP servers; disable broken HTTP MCP plugins to reduce parallel createClient load (helps but not a fix).

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

7

Hey @adiedx-alkami
This is in Cursor’s MCP client orchestration between the app window and the extension host, not your server. That’s why the server keeps serving tool calls on the same PID while the client refresh times out. It’s an issue we’re already tracking.

Best mitigations for now (the ones you found): keep the number of enabled MCP servers as low as you can, and when one gets stuck, toggle it off/on in Settings → MCP or run Developer: Reload Window.

I’ll follow up here once there’s an update.