run_ga4_report — dimensions and metrics parameters fail Pydantic validation when passed as JSON arrays

[sg-modlab]

So odd but I am reporting what Claude is sharing is an issue. I have been having trouble getting to share back some data that requires run_ga4_report. Continues to complain about parameters. I tried with the latest version of AdLoop released today and because this is in Cowork that I experienced this, I copied to the rules shared to an AdLoop skill. So very curious if indeed this is an issue and hope I'm not just making noise here.

---

Tool: mcp__adloop__run_ga4_report

Description:
The dimensions and metrics parameters are documented as accepting lists, but any value passed — including valid JSON arrays — is received by Pydantic as a string and fails validation with Input should be a valid list.

Error:

2 validation errors for call[run_ga4_report]
dimensions
  Input should be a valid list [type=list_type, input_value='["pagePath"]', input_type=str]
metrics
  Input should be a valid list [type=list_type, input_value='["sessions", "totalUsers", "screenPageViews", "bounceRate", "conversions"]', input_type=str]

What was tried:

• ["pagePath"] — fails
• ["sessions", "totalUsers", ...] — fails
• Bare string pagePath — fails with same error

Tool schema (as returned by the MCP server):

{
  "dimensions": { "default": null },
  "metrics": { "default": null }
}

No type is declared for either parameter in the schema, but the underlying Pydantic model expects a list type.

Expected behavior: Passing ["pagePath"] as a JSON array should be accepted and parsed correctly.

Workaround: landing_page_analysis works for paid-session page data, but doesn't support all-source session breakdowns or custom dimension/metric combinations.

[philb74]

I've got the same experience consistently (only in Cowork).
Here's what Claude suggests.

Bug: run_ga4_report — dimensions and metrics parameters fail with list validation error

When calling run_ga4_report via the MCP protocol, passing array values for dimensions and metrics always results in a Pydantic validation error:

Input should be a valid list [type=list_type, input_value='["sessionSource", "sessionMedium"]', input_type=str]

The tool receives the values as str instead of list, regardless of how they are formatted (JSON array, comma-separated string, etc.). This means the tool is currently unusable — passing null triggers "At least one dimension or metric must be specified", and passing any array value triggers the list validation error.

Root cause (likely)

The MCP server's input parsing does not deserialize JSON arrays into Python lists before Pydantic validation runs. The raw JSON string is passed directly to the model instead of being parsed first.

Suggested fix

In the server-side handler for run_ga4_report, add a pre-processing step that parses dimensions and metrics from a JSON string to a Python list if the incoming value is of type str:

import json

if isinstance(dimensions, str):
    dimensions = json.loads(dimensions)
if isinstance(metrics, str):
    metrics = json.loads(metrics)

Environment

MCP protocol. Health check returns ga4: ok, so the underlying API connection is fine.