Connect Your Ad Accounts to Claude Code (2026 Setup Guide)

If you want to connect ads to Claude Code, you are in the right place. People arrive at this question from every direction: "connect ad account to Claude Code", "Google Ads Claude Code", "TikTok ads Claude Code", even niche platforms like AdSwizz or Adalyser. The underlying intent is the same. You live in the terminal, Claude Code is already doing real work for you, and you want your advertising data and your ad accounts wired into it so the assistant can research, diagnose, and launch without you switching to a browser dashboard.

This guide covers the whole thing: why Claude Code is a genuinely good surface for running ads, the exact commands to connect the AdWhispr MCP server, how to link your Google Ads and TikTok accounts, real prompts to try, and an honest section on what to do if your platform is not one we cover.

Why Claude Code is a serious surface for ads

This is not a gimmick. Three properties make Claude Code better at ad operations than most dashboards:

• It is terminal-native. The same session that edits your landing page can pull the competitor ads that page is competing against. Ad context and code context stop being separate tabs.
• It is scriptable. Anything you can prompt, you can wrap in a repeatable workflow. A weekly "check my keyword diagnostics and summarize what changed" routine is a saved prompt, not a standing calendar reminder to go click through reports.
• It runs agentic loops. Claude Code does not just answer one question. It can research a competitor, compare their longest-running ads, pull keyword volumes, and assemble a campaign draft in one continuous chain of tool calls, asking you only when a decision actually needs you.

The missing piece is data and execution access. That is what MCP servers are for, and it is why we built one.

What the AdWhispr MCP gives Claude Code

AdWhispr is one MCP server that covers both halves of the job:

Research (no ad account needed):

• Competitor Meta ad libraries with days-running history. Meta's API returns no run-time data, so AdWhispr builds it from daily snapshots. An ad live 100+ days is a proven winner, because no brand keeps paying to run a loser.
• TikTok ad research via research_tiktok_ads.
• Keyword research: research_keywords for volumes and research_competitor_keywords to see what rivals bid on.
• AI classification of every ad by hook, format, tone, and offer, so questions get structured answers instead of vibes.

Execution (with your own accounts connected):

• Google Search and Performance Max campaign launch, generally available.
• TikTok campaign launch with a full video-ad builder, generally available.
• Meta launch, currently in beta.
• Budget changes, pause and resume, and account performance reads with keyword-level diagnostics.

Competitor data is read-only public Ad Library data. Your own accounts connect through standard OAuth, covered below.

How to connect your ad accounts to Claude Code

The whole setup is two commands plus one OAuth approval.

Step 1: Run the config helper

From any terminal:

npx adwhispr-mcp-server config

The npm package is unscoped: it is adwhispr-mcp-server, not @adwhispr/anything. The helper prints the MCP server entry your client expects, pointing at the hosted endpoint https://adwhispr.com/api/mcp. You will authenticate with your AdWhispr account on first connect. The free tier is enough to confirm everything works.

Step 2: Register the server with Claude Code

claude mcp add adwhispr -- npx adwhispr-mcp-server

Or, if you prefer the hosted HTTP endpoint directly:

claude mcp add --transport http adwhispr https://adwhispr.com/api/mcp

Restart your Claude Code session so it picks up the new server, then confirm:

claude mcp list

You should see adwhispr listed and connected. The full Claude Code setup post walks through the same steps with troubleshooting for tools not appearing, auth loops, and rate limits.

Step 3: Link your Google Ads account

Research works immediately. To launch or read your own performance, link the account. In Claude Code, just ask:

Connect my Google Ads account.

That runs connect_ad_account, which returns an authorization link. Open it, sign in with the Google account that has access to your ads, and approve. This is standard OAuth: AdWhispr gets scoped access to the ad account, never your Google password, and you can revoke it any time from your Google account's connected-apps settings. Then confirm what you are pointed at:

Which ad accounts can I launch into?

list_ad_accounts shows every account the connection can reach. If you manage several accounts under one login, this is where you confirm you are aiming at the right one. There is a browser-client version of this walkthrough in Connect Google Ads to Claude, and the how-to hub has the short-form guides for every platform.

Step 4: Link TikTok

Same pattern:

Connect my TikTok ad account.

One caveat that catches almost everyone: TikTok requires the ad account to be bound to a Business Center, with an identity and a payment method attached, before anything can launch, even a campaign you intend to leave paused. If you have never launched a TikTok ad manually, sort the Business Center out first. It is the most common reason a connection succeeds but a launch fails.

Six prompts to try first

Once claude mcp list shows the server connected, run these in order. They mirror the actual research-to-launch loop:

1. "Use the adwhispr tools to add [competitor] and show me their ad library." Kicks off ingestion of a brand's entire public Meta ad library.
2. "Show me their longest-running active ads and break down the hook, format, and offer on the top three." Days-running is the performance proxy that public data actually supports.
3. "What keywords is [competitor] bidding on? Pick the five that fit my brand best and show me search volumes." Research before spend, every time.
4. "Check my account performance for the last 30 days and tell me which keywords are dragging." With Google linked, this surfaces keyword-level diagnostics such as first-page bid estimates and low quality scores.
5. "Launch a Google Search campaign on those five keywords at $40 a day." Claude Code shows you the assembled campaign, budget, geography, and keywords before it goes live.
6. "Pause the campaign and summarize what we learned this week." Campaign ops as conversation.

The full loop, including cloning a proven competitor creative before launch, is written up in How to Run Ads from Claude.

What if your platform isn't covered?

Honest section. AdWhispr covers Google Ads (Search and Performance Max) for launch and management, TikTok for research and launch, and Meta for research, with Meta launch in beta. That is the list.

Searches also land here for platforms we do not support, so to be clear about the gaps:

• AdSwizz and other audio ad platforms: not supported, and no MCP from us is coming for audio in the near term.
• Adalyser and TV analytics: not supported.
• X Ads: not supported.

If you need one of those inside Claude Code today, you have two generic options, and they are unglamorous but real. First, if the platform has an API, Claude Code can write and run scripts against it directly; you bring your own API credentials and the assistant handles the request code. Second, most of these platforms export CSV. Drop the export into your repo and Claude Code will happily analyze it, join it against other data, and chart it. You lose the live connection and the launch capability, but for reporting and analysis it works today with zero new tooling.

The safety note

Nothing spends money without an explicit instruction from you. Research tools are read-only. Launch tools only fire when you ask for a launch, you set the daily budget in the instruction itself, and you see the assembled campaign before it goes live. Budgets, pausing, and resuming all follow the same rule: the assistant executes what you asked for, and only that. If you ever want out, revoking OAuth access from the platform side immediately stops any further campaign actions.

FAQ

Can Claude Code run my ad campaigns?

Yes, for real. With an ad account connected, Claude Code can launch Google Search and Performance Max campaigns, launch TikTok campaigns including the video ad, change budgets, pause, resume, and read performance. Meta launch is in beta. Every launch requires your explicit instruction and shows you what it built before anything spends.

Does this work in Cursor or Codex too?

Yes. The MCP server is client-agnostic; only the config syntax differs. Cursor has its own walkthrough in the Cursor setup guide, and Codex uses a TOML config block covered in the Claude Code and Codex setup post. The prompts and tools are identical in every client.

Do I need to connect an ad account just to do research?

No. Competitor research, days-running history, keyword research, and TikTok ad research all run on public data with no account connection. Connecting your own account is only required to launch campaigns or read your own performance.

Will it tell me a competitor's CTR or ROAS?

No, because it cannot know them. The Meta Ad Library does not expose CTR, CPC, conversions, or ROAS for competitor ads; those numbers live only inside the advertiser's own account. AdWhispr surfaces what public data actually supports, led by days-running, and any tool that quotes a competitor's exact ROAS is making it up.

One command to start

The whole thing begins with:

npx adwhispr-mcp-server config

Two minutes later Claude Code can read the ad market, and with one OAuth approval it can act on your accounts too. Start free at adwhispr.com.

Your terminal already ships the landing pages. Now it can run the ads that point at them.