Loading ToolBench

Reading the current corpus…

google-ads-transparency-mcp MCP server · B · ToolBench

google-ads-transparency-mcp

github.com/block-town/google-ads-transparency-mcp

Look up any advertiser's Google ads — search by domain or keyword, retrieve ad creatives, decode text ad content, compare active ad counts across competitors and summarise an advertiser's format mix. No API key required.

2026-07-28+STDIO8 tools

73/100

Definition Quality85

Protocol Readiness42

SupportabilityNot measured-

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

This MCP server demonstrates solid definition quality with clear naming conventions, comprehensive parameter schemas, and well-documented tool purposes. All 8 tools follow verb_noun patterns (search_*, get_*, compare_*, fetch_*, download_*) and include detailed descriptions. However, output schemas are documented only in docstrings rather than formal JSON Schema definitions, and some parameter descriptions could better follow the constraint-first pattern (e.g., replacing examples with enums). Error handling guidance is minimal, tools lack recovery hints for common failure modes like rate limiting or missing advertisers. Tool composition is strong: each tool has a single clear responsibility, and the 8-tool set covers a logical progression from search to detail to summary to bulk download. The server correctly exposes human-friendly identifiers (domain, advertiser name) alongside opaque IDs, reducing lookup friction.

Tools (8)

advertiser_summaryread onlysource verified85/100

Summarise one advertiser: format mix, copy sample, destinations. Fetches a sample of the advertiser's creatives and aggregates them: counts and share by format (text/image/video), the last-shown date range, de-duplicated headlines, descriptions and CTA labels, and the hosts the ads send people to. Pure aggregation — read strategy into it yourself.

compare_advertisersread onlysource verified86/100

Compare active Google ad counts across a set of domains, one row each. The competitor paid-surface question in one call: who is running ads, and how many. Rows come back in input order; sort by ad_count yourself. Every row has a status. "ok" means an advertiser was found; "absent" means no advertiser is registered for that domain and IS a genuine zero; "blocked" means Google rate-limited the lookup and is NOT a zero; "error" is anything else. Rows flagged parent_entity resolved to a differently named advertiser (a parent company, a reseller, or a stranger whose ads mention the domain) — never cite those counts as the domain's own.

download_ad_creativeswritesource verified82/100

Download a whole advertiser's creatives into one folder, with a manifest. This is the bulk path for collecting real ad examples — images at their original resolution, videos sampled to stills — so they can be dropped straight into a deck or a report. Each creative gets its own subfolder.

fetch_ad_creativewritesource verified81/100

Download an ad's actual creative to local files. Image ads are always saved at their original resolution, so the file you get back is the real creative — usable in a deck or a slide as-is. Video ads are downloaded and sampled into evenly spaced JPEG frames.

Issues (5)

HIGHtool-schema ↗

Output schemas documented only in docstrings, not as formal JSON Schema. Tool return types are inferred from docstring prose rather than structured schema definitions that clients can parse and validate.

search_advertiser_by_domainsearch_advertisersget_adsget_ad_detailcompare_advertisersadvertiser_summaryfetch_ad_creativedownload_ad_creatives

HIGHrecovery-guide ↗

Error handling lacks recovery guidance. Tools do not document what errors can occur (rate limiting, network failure, advertiser not found) or suggest corrective actions. E.g., compare_advertisers returns 'blocked' status but does not tell the LLM how to retry or interpret that state.

search_advertiser_by_domainsearch_advertiserscompare_advertisers

MEDIUMtool-description ↗

Parameter descriptions include example values (e.g., 'e.g. nike.com', 'e.g. CR10813648716908961793') which LLMs may reuse literally in real calls rather than adapting to context.

Recommendations

Add formal JSON Schema definitions for tool return types. Instead of 'Returns: {domain, status, advertiser_id, ...}' in the docstring, define a returnSchema property with types and descriptions for each field.
Document all possible error states and recovery actions in tool descriptions. E.g., for compare_advertisers, explain what 'blocked' means, how long to wait before retry, and whether partial results are acceptable.
Remove example values from parameter descriptions and replace with format constraints. Change 'domain (e.g. "nike.com", "coinbase.com")' to 'domain: a valid DNS hostname (e.g., lowercase alphanumeric and hyphens)' or better yet, add a regex pattern constraint.
Add explicit min/max bounds to numeric parameters in descriptions. E.g., 'count: Number of ads to retrieve (1 - 100, default 10)' instead of relying on implementation validation.
Register tools with readOnlyHint=true for search/get tools and destructiveHint=true for download tools (since they write files). Formally document idempotency via tool annotations.
Add dependency hints to tool descriptions. E.g., 'To get an advertiser_id, first call search_advertiser_by_domain() or search_advertisers().' This prevents wasted lookup calls.
Enhance error responses with actionable next steps. When search_advertiser_by_domain returns None, the error should suggest 'Advertiser not found for domain. Try search_advertisers() with the company name instead.'
Document rate-limiting behavior and backoff strategy. The compare_advertisers tool mentions 'blocked' status due to rate limiting but does not explain the retry policy.

Score history

Overall score trend

First recorded score · v2 rubric

73/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	B	73	2026-07-28+	v2