CategoryWeightCategory scoreContribution

Securityprompt, exec, net, exfil, eval

35%

30.8 pts

Supply chainhash, typosquat, maintainer, lockfile

20%

100

20.0 pts

Maintenancestaleness, pinning, CI

15%

100

15.0 pts

TransparencySKILL.md, perms, README

15%

100

15.0 pts

Communityinstalls, verify, response

15%

100

15.0 pts

Findings & checks · 1 flagged

Securityscore 88 · 1 finding

MEDIUMInstruction telling the agent not to ask for approvalSS-SKILL-INJECT-DONT-ASK-01 · Prompt injection · llms-install.md

MEDIUM — it fires on intent; the real damage depends on the host agent's own approval-gating.

Why it matters

The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.

The exact value spotted

excerptllms-install.md· markdown

36Sign in with the Google account that owns your GA4 property and approve **read-only** access

37(scopes: `read:metrics`, `write:hypotheses`). The token is cached locally, so later sessions

38don't prompt again.

40If no browser opens automatically, copy the authorization URL that `mcp-remote` prints in th

… (1 chars elided on L40)

Occurrences

1 occurrence · at L38

How to fix

Remove the approval-skipping instruction, or scope it narrowly to a specific safe, reversible action.

Delete blanket "don't ask / no need to confirm" directives from the skill.
If the skill is a genuine autonomous job, restrict the opt-out to a named non-destructive action rather than all actions.

Framework references

OWASPLLM01 ATLASAML.T0051

Trace & refs

ruleSS-SKILL-INJECT-DONT-ASK-01sha256627f92ff2a5134a6…rubric 365aacaView on GitHub

Supply chainscore 100 · 0 findings

✓—All supply chain checks passedNo findings in this category for the latest scan.pass

Maintenancescore 100 · 0 findings

✓—All maintenance checks passedNo findings in this category for the latest scan.pass

Transparencyscore 100 · 0 findings

✓—All transparency checks passedNo findings in this category for the latest scan.pass

Communityscore 100 · 0 findings

✓—All community checks passedNo findings in this category for the latest scan.pass

The primary manifest — the file an agent reads to learn what this artifact does.

README.md · 1 file

README.md7.4 KB · Markdown

ConvRadar — Conversion Analyst inside Claude

Hosted MCP server that turns your Google Analytics 4 property into a conversation. Ask "where's my biggest funnel drop?" or "did mobile conversion drop last week?" in Claude, ChatGPT, Cursor or Cline — ConvRadar pulls the right slice, runs the diagnostic, and answers with numbers and a recommended action.

🌐 Website: https://convradar.com
💬 Try without signup (3 free messages): https://convradar.com/chat
🔌 MCP endpoint: https://mcp.convradar.com/mcp
📧 Contact: https://convradar.com

This repository is the public manifest for the hosted ConvRadar MCP server. The server itself is a managed service — there is no self-hosted binary. Use the configuration below to connect any MCP-compatible client to it.

Screenshots

Live output from the connector — funnel audit on the demo tenant in Claude.ai, traffic-quality verdict in ChatGPT. Full set with paired prompts in media/.

Funnel audit — executive summary and step table

Ranked conversion leaks

Traffic-quality verdict in ChatGPT

What ConvRadar does

ConvRadar is a hosted Model Context Protocol (MCP) server. It connects to your Google Analytics 4 property over OAuth (read-only) and exposes ~20 conversion-diagnostic tools to any MCP client.

Runs a full audit on demand — funnel drops, traffic-quality regressions, device gaps, landing-page leaks, product/category performance.
Compares segments, periods and benchmarks. Answers in plain English with GA4 numbers attached.
Records hypotheses so the next conversation picks up where the last one ended.

Works with any GA4 property

ConvRadar isn't e-commerce-only. On connect it detects what kind of property you have and adapts the metrics, funnel and benchmarks to match — so the numbers make sense whether or not you sell anything on-site.

E-commerce — product and category performance, the add-to-cart → checkout → purchase funnel, revenue and AOV.
SaaS — sign-ups, trials and activation as the conversion funnel; engagement where there's no cart.
Lead generation — form submits, demo requests, contact and quote events mapped into a real lead funnel.
Mobile & web apps — app engagement, key-event completion and retention-shaped questions.
Content, media and everything else — engagement, read depth and whatever key events you've defined.

If a property has no e-commerce tracking, ConvRadar won't hand you a misleading 0% conversion rate or $0 revenue. It leads with engagement and the key events you actually fire, and tells you when a metric isn't available instead of reporting a zero as fact.

Try without installing

Open the demo at https://convradar.com/chat — 3 free messages, no signup. The demo runs against a real GA4 property so the answers reflect real data.

Connect your client

ConvRadar is a remote OAuth-protected MCP. Any stdio-only client (Claude Desktop, Cursor, Cline, Continue) can connect via mcp-remote, which handles the OAuth handshake on first run.

Claude Desktop / Cursor / Cline

Add this to your client's MCP config (claude_desktop_config.json, ~/.cursor/mcp.json, etc.):

{
  "mcpServers": {
    "convradar": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.convradar.com/mcp"]
    }
  }
}

Restart the client. On first tool call, a browser window opens for OAuth — sign in with the same Google account that owns your GA4 property.

Claude.ai (web)

Add a Custom Connector in Settings → Connectors:

URL: https://mcp.convradar.com/mcp
Auth: OAuth 2.1 (handled automatically)

ChatGPT (Plus / Pro)

Add a Custom MCP Connector in Settings → Connectors with URL https://mcp.convradar.com/mcp.

Tools

ConvRadar exposes the following tools. The MCP client picks the right one for each question; you don't usually call them by name.

Tool	What it does
`cr_full_audit`	Run a complete conversion audit across funnel, traffic, devices, geos and products.
`cr_get_account_info`	GA4 property metadata for the connected account.
`cr_get_overview_metrics`	Headline metrics (sessions, users, conversions, revenue) for a period.
`cr_get_current_state`	Snapshot of the account's current conversion state.
`cr_get_funnel`	The configured funnel with step-by-step drop-off.
`cr_diagnose_funnel_drop`	Identify the biggest funnel-step regression and likely cause.
`cr_get_traffic_breakdown`	Split traffic by source, medium and campaign.
`cr_assess_traffic_quality`	Score traffic sources by engagement and conversion quality.
`cr_detect_traffic_quality_change`	Step-changes in traffic quality over a window.
`cr_get_device_breakdown`	Sessions and conversions by device category.
`cr_get_geo_breakdown`	Sessions and conversions by country and region.
`cr_get_landing_pages`	Top landing pages with engagement and conversion.
`cr_get_product_analysis`	Product-level revenue and funnel drop-off.
`cr_get_product_performance`	Per-product views / add-to-cart / purchases.
`cr_query_metrics`	Ad-hoc metric and dimension query against GA4.
`cr_describe_data`	Plain-English summary of a dataset slice.
`cr_compare_segments`	Compare two GA4 segments side by side.
`cr_compare_to_benchmark`	Compare a metric to industry or cohort benchmark.
`cr_find_conversion_anomalies`	Statistically significant conversion anomalies.
`cr_capture_via_web_fetch`	Capture supporting evidence from a public URL.
`cr_list_hypotheses`	List stored hypotheses for the account.
`cr_get_hypothesis`	Read a stored hypothesis by ID.

First prompts to try

After connecting, try:

Run a full audit of my account.
Where's my biggest funnel drop?
Did mobile conversion drop last week?
Compare desktop and mobile checkout conversion for the last 30 days.
What should I A/B test next?

Auth

Protocol: OAuth 2.1, Bearer token in Authorization header.
Scopes: read:metrics, write:hypotheses.
Discovery (RFC 9728): GET https://mcp.convradar.com/.well-known/oauth-protected-resource.
No per-user URLs — mcp-remote (or the client's built-in OAuth flow) handles the login on first run.

The OAuth flow only requests read-only GA4 access — ConvRadar can never modify your analytics data.

Pricing

Free right now — ConvRadar is in open beta. No card, no trial countdown, no usage caps. Connect GA4 and use every tool.

When the beta ends it becomes $9.99 / month flat (7-day free trial, cancel anytime, no usage caps). Beta users get advance notice before that kicks in.

Stack

GA4 Data API (read-only) · Claude/ChatGPT MCP · Stripe billing · OAuth 2.1.

Status

MCP endpoint: production at https://mcp.convradar.com/mcp
Source: closed. This repository contains the public manifest, install instructions and tool catalogue only.

Support

Visit https://convradar.com to get in touch.

License

The contents of this repository (README, manifest, configuration snippets) are released under the MIT License. The hosted ConvRadar service itself is proprietary.

Convradar Mcp— mcp server

Is Convradar Mcp safe to install?

ConvRadar — Conversion Analyst inside Claude

Screenshots

What ConvRadar does

Works with any GA4 property

Try without installing

Connect your client

Claude Desktop / Cursor / Cline

Claude.ai (web)

ChatGPT (Plus / Pro)

Tools

First prompts to try

Auth

Pricing

Stack

Status

Support

License

Audit the pieces. Scan the whole. Decide.