Does Claude spend money without asking?

No. Read-only requests run freely, but any campaign create, edit, or budget change returns a preview you must explicitly approve. The preview is single-use, so you always confirm the exact plan that gets submitted.

Can I manage multiple LinkedIn ad accounts?

Yes. You authorize each ad account you want managed, and you can ask Claude to list accessible accounts and target a specific one for any action. Access is scoped to what you granted and revocable from LinkedIn.

Do I need to be a developer to use it?

No. After the one-time OAuth connection, everything is done in plain language. Knowing LinkedIn's campaign-group / campaign / creative structure helps you phrase requests, but you never write code.

How to Run LinkedIn Ads From Claude

VigilDog Team · September 20, 2026 · 6 min read

You can now run LinkedIn Ads from Claude, describe the campaign you want in plain English and an approval-gated assistant builds it against the real LinkedIn Marketing API. No Campaign Manager tab-hopping, no bulk CSV sheets. Here is how the workflow actually works, and where you should keep your hands on the wheel.

What "running LinkedIn Ads from Claude" actually means

The VigilDog Ads MCP is a Model Context Protocol server that connects Claude (or ChatGPT) to LinkedIn's advertising API. Claude is handed a set of typed tools, list ad accounts, create campaign groups, define targeting, build lead-gen forms, upload creatives, pull analytics, and it calls those tools on your behalf when you ask for something in chat.

Nothing here bypasses LinkedIn's object model, so it helps to know it. An ad account contains Campaign Groups, which contain Campaigns, which hold Creatives. Objective, audience, budget, schedule, and bid all live at the campaign level. When you say "launch a lead-gen campaign for our webinar," Claude maps that sentence onto exactly these objects.

If you have already used the same approach for search, it will feel familiar, the pattern mirrors how you run Google Ads from Claude, just against a different network's rules and inventory.

Connect your LinkedIn ad account

Before Claude can touch anything, you authorize it once. You sign in with LinkedIn, grant the specific ad account you want managed, and the connector stores the resulting OAuth token encrypted. Claude never sees your password, and you can revoke access from LinkedIn at any time.

A typical first-run looks like this:

Open the Ads MCP connector and start the LinkedIn authorization flow.
Sign in and approve the ad account (and only that account) you want to manage.
Ask Claude to list your accounts to confirm the connection is live.
Pull a quick performance snapshot so you know the baseline before you change anything.

How an Ads MCP runs a change safely

Launch a campaign by describing it

Once connected, you drive everything with prose. A real prompt might be: "Create a lead-gen campaign in the Demand group, targeting Heads of Marketing and CMOs at 200+ employee SaaS companies in the US, single-image ads, 60 dollars a day, max-delivery bidding, running for four weeks."

Claude translates that into concrete API calls. Under the hood it is choosing an objective (lead generation, website visits, engagement, video views, or brand awareness), assembling an audience from LinkedIn's targeting facets, setting a budget above LinkedIn's per-campaign daily minimum, picking a bid strategy, and attaching the right creative format.

The pieces it commonly wires up for you:

Targeting facets: job title, seniority, function, industry, company size, named company lists, member skills, and groups.
Creative formats: single image, carousel, video, document, thought-leader, and conversation/message ads.
Budget and bid: daily or lifetime budget with max-delivery, cost-cap, or manual bidding.
Lead-gen forms: prebuilt question sets so responses flow straight into your CRM.

The guardrails: dry-run and approval

The point that makes this safe to use on a live account is that writes are not fire-and-forget. When you ask Claude to create or change something that spends money, it first returns a preview, the exact objects, targeting, and budget it intends to submit, and waits for your explicit confirmation.

That preview is single-use and tamper-checked, so you cannot accidentally approve a stale plan, and a policy guardrail flags obvious problems (missing creative, audience too narrow, budget below the network minimum) before anything reaches LinkedIn. Read-only questions like "how did last week perform?" run freely; anything that mutates the account stops at the approval gate.

What it is great at, and what to still do yourself

Where this shines is the mechanical work that eats an afternoon: scaffolding a campaign group, cloning a structure across regions, bulk-editing bids, building a lead form, and pulling analytics into a readable summary. Iterating ad copy against your own brief is fast because you are editing in conversation, not in twelve modal dialogs.

What it will not do for you is strategy. The offer, the creative concept, the final sign-off on brand voice, and the authority over budget stay with you, and in regulated industries you still own compliance review. Treat Claude as a very fast media buyer who always asks permission before spending, not as an autopilot.