AdRadar / Features / MCP server
Features · MCP server
Ask your AI assistant about competitors. It reads AdRadar.
AdRadar works as an MCP server. Connect Claude, Claude Code, Cursor or ChatGPT once, then ask in words: what did a competitor launch this week, what do its customers complain about, who bids on our brand. The assistant pulls the data from AdRadar and answers with links to the ads, pages and reviews behind it.
Who is most active this week, and who is losing ground?
Claude is the most active: 483 new ads this week, and 5 of the 7 companies launched ads too.
- Lovable57 new ads, 1,665 active, 332 running 90+ days. 9 new site pages this week.
- vidIQ73 new ads and 1,525 new site pages in a week: a content push, not just ads.
- Replit13 new ads, 3 stopped. Only 4 ads have run 90+ days: the creatives rotate fast.
- TubeBuddy, Celloone new ad each. Nothing new from the rest.
Questions in words. Tools behind them.
You never name a tool. The assistant picks it, calls AdRadar and writes the answer with links to the evidence.
What did Lovable do this week?
Headline, trend per channel, new and stopped ads, new site pages.
Who is most active this month? Who is losing ground?
One line per company and one sentence about the market.
Write hooks for DE from what they have run 60+ days
Filtered by country and days running, sorted by the longest run. Then the assistant writes.
What do their customers complain about that we solve?
Rating per site, the words that mark bad reviews, quotes from 1–2★ originals.
Has anyone changed prices?
Price changes on landing pages, with the page and the old and new price.
Who bids on our brand in France?
Advertisers on your brand and key queries, by country.
What do they publish, where does traffic come from?
New pages by type, organic traffic by country, the pages that lead.
Show me this ad in full
Text, video transcript, text from the image, when it was seen active.
Three steps, no key.
01
Add the address
In claude.ai, ChatGPT, Slack, or in Cursor, Claude Desktop or Claude Code, add https://app.adradar.digital/api/mcp as a connector or remote MCP server.
02
Sign in
Server address https://app.adradar.digital/api/mcp. The client opens AdRadar’s sign-in. Allow access and the assistant reads your organization. For scripts, a key from Settings → API keys works instead.
03
Ask
“Which companies do we track and what is known about each?” Then “What changed this week, without reviews?”
$ claude mcp add --transport http adradar \
https://app.adradar.digital/api/mcp
Added HTTP MCP server adradar
$ claude
› /mcp → adradar · sign in (opens the browser once)
› adradar · 13 tools · connected
› What changed at our competitors this week, without reviews?| Where | How |
|---|---|
| Claude Code | The command above, then /mcp once to sign in. The 13 tools appear. |
| Cursor Claude Desktop | Remote server, type http, https://app.adradar.digital/api/mcp, header Authorization: Bearer <key>. |
| claude.ai Claude in Slack ChatGPT | Settings → Connectors → Add custom connector: https://app.adradar.digital/api/mcp, then sign in when asked. |
A client that cannot sign in takes the key in a header, or in the address (https://app.adradar.digital/api/mcp/<key>, as secret as the key). Full reference: docs.
One question instead of 20 minutes in a dashboard.
A look at a dashboard changes nothing. Attach the question to a ritual your team already has.
“Read what changed last week for project X and write three things worth deciding.”
A scheduled task in Claude. It arrives at the planning meeting, not when someone remembers.
“Which angles and hooks have competitors kept live for 60+ days in DE that we have not tried?”
The creative sprint starts from what already works, with a link to every ad.
“What do X’s customers write about prices and support in the last 30 days, with quotes.”
Client prep from real reviews, not memory.
“Who raised a price, who started bidding on our brand, who is most active in FR?”
The founder’s question, answered in one message with the evidence linked.
Thirteen tools, all read-only.
The assistant takes ids from list_projects and list_competitors and passes them on. Nothing here writes: it cannot add a company, change an alert or delete anything.
Start here
list_projects
Projects and how many companies each tracks.list_competitors
Companies, the state of each source, last check.compare_competitors
One line per company, one sentence about the market.list_reports
The weekly AI reports of a project.
Ads
list_ads
Ads with AI labels. Filter by company, platform, status, format, country, angle, days running, text.get_ad
One ad in full: text, transcript, text from the image, media.recent_changes
Events since a date as sentences: new and stopped ads, bursts, proven creatives, price changes, new pages, bad reviews.
Reviews
Capterrareview_insights
Per review site: rating, counts, the words that mark bad and good reviews, the AI digest with quotes.list_reviews
The reviews themselves and Reddit mentions.
Site and search
landing_pages
Pages ads lead to, with the price found on them.site_changes
Pages by type, additions and removals, organic traffic by market, leading pages.brand_queries
Who buys ads on your brand and key queries, by country.
Counts with a date, not forecasts.
Everything the assistant says comes from data AdRadar holds, and links back to the ads, pages and reviews behind it.The assistant gets these rules when it connects. They are worth knowing as a person too.
For people who ask, not browse.
Founders and vibe coders
“What did the companies I compete with launch this week?”
You already live in Claude Code or Cursor. Ask without opening another tab.
Performance marketers
“Give me the 10 longest-running Meta hooks in our market, with links.”
Straight into the brief, with a link to every ad.
Agencies
“Monday summary for client X, three decisions, to Slack.”
One key per client organization. The assistant writes the summary, Slack delivers it.
Questions
What is an MCP server?
MCP (Model Context Protocol) is an open standard that lets an AI assistant call tools in another product. AdRadar exposes its competitor data as 13 read-only tools. Claude, Cursor, Claude Code or ChatGPT connect to them once, and from then on you ask in words and the assistant decides which tool to call.
Which assistants work with it?
Claude Code, Cursor, Claude Desktop and any client that supports remote HTTP MCP servers with a header, plus claude.ai, Claude in Slack and ChatGPT through a custom connector URL.
Can the assistant change anything in AdRadar?
No. Every tool only reads. It cannot add a competitor, change alerts or delete anything, and it sees only the organization the key belongs to.
Which plan do I need?
Every plan, Free included: the MCP server is open to all organizations, and an assistant signs in without a key. API keys (for scripts and the REST API) are made on Settings → API keys; the REST API itself is part of Pro and Agency. 60 requests a minute per key or per signed-in person.
Where does the data come from?
From what AdRadar has collected for your organization: the public ad libraries of Meta, Google, LinkedIn and TikTok, the landing pages behind the ads, G2, Capterra, Trustpilot and Reddit, sitemaps and organic traffic estimates. The assistant gets no forecasts, budgets or spend figures, because AdRadar has none.
Can the assistant make numbers up?
It answers from tool results. Every number is a count of what AdRadar holds, every event is something observed between two checks, and review digests cite the review ids they rest on. Ask it to show the ads or reviews behind a claim and it will link them.
Is the connector URL with the key safe to share?
Treat it like the key itself: whoever has it can read your data. Revoke the key on the API page and both the header and the URL stop working at once.
Can I schedule questions?
Yes, in any assistant that runs scheduled tasks, for example Claude. A common one: every Monday, read what changed in the last week and list three things worth deciding.