Loading ToolBench
A Cloud-Agnostic, First-Principles Model Context Protocol (MCP) server for Google Meridian. Provides documentation retrieval, mathematical modeling tools, workflow guidance, and code synthesis for media mix modeling (MMM) and marketing attribution analysis.
Server has 11 tools with mostly complete schemas and descriptions, but quality is inconsistent. Tool names follow verb_noun convention well (list_, fetch_, search_, get_, calculate_, audit_, synthesize_, generate_, run_). Descriptions range from adequate (194 chars avg baseline) to good, but several lack actionable context for LLM selection. Parameters are typed and mostly described, but many lack constraints (enums, ranges, format specs). Output schemas are not documented, LLMs cannot predict response structure. Error handling is minimal; tools return generic exceptions rather than recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all being read-only operations. Security is reasonable (SSRF checks, URL allowlisting, no secrets in params), but audit logging is basic.
Audits a proposed Google Meridian model specification for identifiability, knot density, and bounds.
Solves exact probability equations to convert 95% CIs into Meridian LogNormal (mu, sigma) prior parameters.
Fetch and parse Google Meridian documentation or code into clean Markdown.
Generates a synthetic benchmarking CSV schema template matching Meridian's input format.
Provides operational parameters, mathematical formulas, recommended bounds, and Meridian defaults for data science control points.
Provides step-by-step decision trees and iteration rules for the 5 modeling phases and 3 iteration loops.
Output schemas not documented. LLMs cannot predict response structure for any tool. Tools return free-form strings or unstructured markdown, forcing LLMs to parse unstructured output and plan downstream calls blindly.
No tool annotations despite all 11 tools being read-only. Missing readOnlyHint declarations prevent agents from understanding which tools are safe to call speculatively and which require confirmation.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | <=2025-11-25 | v2 |
List all available Google Meridian documentation sources, user guides, API references, and GitHub repository links.
Provides guidelines and configuration for Google Meridian Pre-Modeling EDA checks (VIF, Pairwise Correlation, KPI Variance, Data-Parameter Ratio).
Provides diagnostic checks from Google Meridian Automated Model Review Engine (meridian.analysis.review.checks).
Search Google Meridian documentation by topic or keyword (e.g., 'adstock', 'priors', 'budget optimizer', 'NUTS', 'R-hat', 'DAG', 'causal').
Synthesizes clean, portable, cloud-agnostic Python code implementing Google Meridian pipelines.
Parameter constraints missing. 'category' in list_doc_sources accepts free-form strings but should be an enum. 'check_type' in run_model_review_checks lists valid values in description but not as JSON Schema enum, inviting hallucinated values.
Error handling lacks recovery guidance. fetch_docs returns 'Error: URL not from allowed domain' but does not suggest alternatives or next steps. No categorization of errors as retryable vs user-fixable.
Descriptions for audit_model_first_principles, synthesize_meridian_code, and run_eda_checks are vague. 'Audits a proposed model' does not explain WHAT is audited (identifiability? bounds? knot density?) or WHEN to call it vs other tools.