Looker-toolbox MCP client times out on all calls despite endpoint being healthy

Support Bug Reports
, ,
1

Where does the bug appear (feature/product)?

Cloud Agent (GitHub, Slack, Web, Linear)

Describe the Bug

looker-toolbox MCP client consistently times out when calling get_models, get_explores, and other tools, with error: “MCP error -32001: Request timed out.”

The endpoint itself is healthy and responsive—direct curl calls to the Looker MCP endpoint succeed in ~367ms. This indicates the bug is in the MCP client library or Cursor’s configuration, not the Looker endpoint.

This blocks all Looker-based Cloud Agent Automations from functioning. They cannot query Looker to answer data questions, build reports, validate LookML, or create PRs.

MCP Server Config (structure):
{
“type”: “http”,
“url”: “https://[company].looker.com/mcp”,
“auth”: {
“CLIENT_ID”: “[redacted]”,
“scopes”: [“cors_api”]
}
}

Environment: Cloud Agent VM (team-managed, no repo environment.json)
Reproduction: Call any looker-toolbox tool (e.g., get_models) → timeout after ~30s
Contrast: Snowflake-DE-MCP-Server works fine in the same environment

Steps to Reproduce

  1. Set up a Cloud Agent Automation with looker-toolbox MCP configured
  2. Trigger the automation to call any looker-toolbox tool (e.g., get_models)
  3. Observe the MCP client hangs and times out after ~30s
  4. Error returned: “MCP error -32001: Request timed out”

To verify endpoint is working:

  • Run: curl -X POST https://[company].looker.com/mcp -H “Content-Type: application/json” -d ‘{“jsonrpc”: “2.0”, “id”: 1, “method”: “tools/list”}’ --max-time 10
  • Result: HTTP 200 in ~367ms (proves endpoint is responsive)

Expected Behavior

The looker-toolbox MCP client should successfully connect to the Looker MCP endpoint and return tool definitions or query results within 1-2 seconds, similar to how other MCPs (e.g., Snowflake-DE-MCP-Server) work in the same Cloud Agent environment.

Operating System

Linux

Version Information

Cloud Agent environment, not IDE/CLI.
This occurs in a Cloud Agent Automation running on a pre-built environment snapshot.
If needed, the issue affects all versions using looker-toolbox MCP.

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

6

Hey @Hayley_Merrigan, thanks for the detailed writeup!

Quick unblock: run Looker as a remote HTTP MCP server instead of stdio. Your instance exposes a hosted endpoint, so adding it under cursor.com/agents → MCP as { "type": "http", "url": "https://<your-instance>.looker.com/mcp" } and authorizing it once makes it run from our side (like Snowflake already does for you), instead of inside the agent VM.

The short why: a stdio server runs inside the VM and reaches Looker over a different network path than your laptop’s curl, so it can hang there and hit the 60s timeout even while the endpoint itself is healthy. HTTP avoids that entirely - it’s the recommended transport (docs).

If you’d rather keep it as stdio, paste the exact looker-toolbox definition you have saved (secrets redacted) and I’ll take another look.