Google运维分析

概述

Google Analytics

Overview

Use the project Google Analytics MCP workflow for GA4 read-only analysis. The Skill is stored in this repository for partner/project use; the MCP server normally runs on the cloud host through SSH stdio and a temporary Docker runner.

Primary project reference:

• docs/GOOGLE_ANALYTICS_MCP.md

Runtime Model

Preferred first-stage runtime:

local MCP client / local Skill
  -> ssh -T cloud host
  -> docker run --rm -i cloudplus/analytics-mcp
  -> cloud-host ADC credential mount
  -> Google Analytics Admin API / Data API

Do not ask users to configure local Google ADC when the project cloud runner is available.

When To Use

Use this Skill for:

• GA4 traffic, acquisition, engagement, retention, or conversion analysis.
• Event popularity, event parameters, custom dimensions, and custom metrics.
• Funnel or path-style questions that can be answered by GA Data API reports.
• Realtime active user questions.
• Account/property discovery and property configuration questions.
• Google Ads link discovery for GA properties.

Do not use this Skill for:

• Editing GA4 configuration.
• Writing credentials into .env, source files, or chat output.
• Non-GA SEO checks such as Google Search Console unless another project doc/tool covers it.

Workflow

1. Read docs/GOOGLE_ANALYTICS_MCP.md if the runtime, credential, or cloud runner setup is relevant.
2. Confirm the MCP server is available in the MCP client. If unavailable, report that analytics-mcp must be configured through SSH stdio.
3. If the GA4 property is ambiguous, call get_account_summaries first.
4. For property metadata, use get_property_details, list_google_ads_links, or get_custom_dimensions_and_metrics.
5. For historical reports, require or clearly infer:
   • property id
   • date range
   • metrics
   • dimensions
   • filters or comparisons
6. For live activity, use run_realtime_report, not historical report tools.
7. Summarize results with the property, date range, dimensions, metrics, filters, and any assumptions.

Defaults

When the user omits details and the request is low risk:

• Date range: use last 28 days for trend/report questions.
• Realtime: use run_realtime_report.
• Property: discover with get_account_summaries rather than guessing.
• Sorting: sort descending by the primary metric.
• Limit: keep result sets small enough to explain clearly.

State all defaults in the final answer.

Report Guidance

Prefer concise, decision-oriented summaries:

• Start with the main finding.
• Include a compact table when comparing dimensions such as page path, event name, country, channel, or campaign.
• Mention whether metrics are users, sessions, event count, conversions, revenue, or another GA metric.
• Avoid overclaiming causality from descriptive GA reports.
• Call out sparse data, sampling/thresholding, missing property access, or unclear date ranges.

Credential Safety

• Never request or expose credential JSON content.
• Never suggest committing ADC files, service-account keys, or OAuth client JSON files.
• Treat SSH aliases, hostnames, credential paths, and property IDs as deployment-sensitive when the user has not already disclosed them.
• If credential setup is needed, point to docs/GOOGLE_ANALYTICS_MCP.md and keep examples placeholder-based.

Partner Project Notes

This repository keeps the Skill source under skills/cloudplus-google-analytics/ for project maintainers and partners. A local Codex installation may need to copy or symlink this folder into its user-level Skill directory before automatic triggering works.