A Claude Meta Ads connector lets Claude read and act on your Meta ad account through the Model Context Protocol (MCP), so you stop exporting reports and pasting them into a chat window. AdStellar provides one at a single MCP URL, and in Claude Code the setup is one command.
The practical difference is freshness and reach. A pasted CSV is a snapshot that starts aging the moment you export it, and Claude can only comment on it. A connected Claude can query current data, reason over it, and, through AdStellar's tools, help build and launch campaigns. The sections below cover how the connection works, how to set it up, what to ask for, and where you should keep your hands on the wheel.
How a Claude Meta Ads Connector Works
MCP, the Model Context Protocol, is an open standard from Anthropic that lets AI assistants call external tools and read external data through a server. Instead of every integration being custom code, an MCP server publishes a list of tools (things the assistant can do, such as "fetch campaign performance") and the assistant decides when to call them during a conversation. Each of those calls is a tool call.
A Meta Ads connector has three parts:
- Claude as the client. It reads your request, decides which tools it needs, and interprets the results.
- The MCP server as the translator. It exposes ad-account capabilities in a form Claude understands and handles authentication and execution.
- Meta Ads as the destination. Campaign, ad set, and creative data, plus the actions that create or change them.
A common misconception is that Claude can reach Meta by itself. It cannot. Out of the box, Claude has no access to your ad account, and no amount of clever prompting changes that. Something has to sit in the middle and expose tools, and that is the role AdStellar's MCP endpoint plays.
The workflow it replaces
The old routine goes like this: open Ads Manager, set a date range, export a CSV, paste it into a chat, ask what is wrong, then repeat tomorrow because yesterday's answer no longer matches the account. Large exports get truncated or lose context, and any recommendation still has to be carried back to Ads Manager and implemented by hand. With a connector, the question and the data live in the same place, and the answer can lead directly to an action.
Why Connect Claude to Meta Ads Through AdStellar
AdStellar is built as an agentic AI media buyer: it analyzes performance, creates ads, and launches campaigns, rather than only reporting on them. Connecting Claude to it means your conversational assistant gets those capabilities as tools instead of you wiring up raw API access yourself.
Several features matter in practice:
- Real-time Meta data. Analysis reflects current spend and performance, not whatever you last pasted.
- Transparent reasoning. Each campaign comes with the logic behind it, so a media buyer can audit why a structure or audience was chosen rather than trusting a black box.
- Element-level creative scoring and fatigue detection. Ads are scored by their components, which helps you see what is actually driving results and when a creative is wearing out.
- Bulk launch. Hundreds of ad variations can go live in minutes, which is where manual workflows usually break down.
- A ranked creative library. Proven winners are stored and can be reused in new campaigns.
- Unlimited workspaces. Agencies can keep each client's account and assets separate.
AdStellar can also build a full data-backed campaign in under 60 seconds from your past performance, and it covers creative production too: image ads, video ads, and AI UGC avatar and talking-head ads.
Who it fits
This is a strong fit for performance marketers, e-commerce brands, and agencies running Meta at volume, where the time cost of analysis, creative production, and launch adds up. If you run one small campaign with a handful of ads, you probably do not need an agentic layer, and Ads Manager alone may be enough. Pricing has ranged from $100 to $10,000 as of this writing, so check the current plans on the AdStellar site before committing.
Set Up the Connection in Claude Code: Step by Step
Before you start, you need two things: an AdStellar account with your Meta ad account already connected, and Claude Code installed on your machine.
- Add the server. In your terminal, run:
claude mcp add --transport http adstellar https://app.adstellar.ai/api/mcp - Complete authorization. If prompted, follow the sign-in flow to link your AdStellar account. The exact prompt can change, so follow what Claude Code and AdStellar display.
- Verify registration. Run claude mcp list in your terminal, or type /mcp inside a Claude Code session, and confirm adstellar appears and shows as connected.
- Test it. Ask a harmless read-only question, such as which campaigns spent the most in the last seven days.
The only MCP URL to use is https://app.adstellar.ai/api/mcp. Do not substitute a third-party address, a guessed variation, or a URL from a forum post. This endpoint will have access to your ad account, so the source matters.
If it does not work
- Typo in the URL. Compare it character by character with the one above.
- Wrong transport flag. The command uses --transport http. Another transport type will not connect to this endpoint.
- Unfinished authorization. The server may be registered but not signed in. Check /mcp and complete the flow.
- No Meta account in AdStellar. The connection can succeed while returning no data if no ad account is linked inside AdStellar.
Claude Code's CLI commands and flags evolve. As of this writing the steps above match its documented MCP workflow, but check Anthropic's current Claude Code documentation if a command behaves differently.
What You Can Ask Claude to Do Once Connected
Think in three categories: analysis, creation and launch, and reuse. The exact tools the server exposes determine what is possible, so run /mcp to see what is listed for your account and treat that list as the source of truth rather than any article, including this one.
Analysis
Ask which ads are showing fatigue, which creative elements correlate with your best results, and where spend is leaking into ad sets that are not converting. Because AdStellar scores creative at the element level, questions about hooks, formats, or visual treatments can be answered against your own account history.
A sample prompt: "Review the last 14 days. Which ad sets are spending without conversions, and which creatives look fatigued? Explain your reasoning before suggesting changes."
Creation and launch
This is where an agentic setup differs from a reporting tool. You can ask for a campaign built from past performance, new image or video ads, or UGC avatar ads, and for variations launched in bulk.
A sample prompt: "Build a prospecting campaign for our best-selling product using what has worked in the last 90 days. Daily budget capped at $150. Show me the structure and reasoning, and wait for my approval before launching."
Reuse and import
Ask Claude to pull top-ranked winners from the creative library into a new campaign, or to import an existing Meta campaign so it can be analyzed and extended.
A sample prompt: "Take our three highest-ranked creatives and set up a retargeting test against past site visitors. Use consistent naming so I can read results by creative."
Attribution alongside platform data
If you use Cometly for attribution, AdStellar integrates with it, which lets you look at attributed results next to what Meta reports. That comparison is useful for the cross-checks covered next.
Guardrails: Keeping a Human in the Loop
Connecting Claude does not mean spend happens without review. Autonomy is something you grant in the prompt and in your process, and it is worth granting in stages.
Constrain before you launch
State budget caps, the campaign goal, target geography, and any exclusions before asking for anything to go live. Ask Claude to present the plan first and wait for your approval. A prompt without constraints invites a plan built on assumptions you never checked.
Audit the reasoning
AdStellar shows the reasoning behind each campaign. Read it. Before launch, confirm the audience, placements, and budget match your intent, and check that the structure suits how you actually evaluate results.
Mistakes to avoid
- Judging ads on too little data. A creative with a few dozen impressions has told you almost nothing. Set minimum spend or conversion thresholds before declaring winners or losers.
- Launching hundreds of variations without a plan. Bulk launch is powerful, but without a naming convention and a clear testing structure you will generate noise you cannot interpret. Decide what each variation is testing first.
- Trusting platform ROAS alone. Meta's reported returns are one view. Cross-check against attribution data, such as Cometly, before scaling budgets.
Earn trust gradually
Start with read-only analysis prompts for a week or two. See whether Claude's findings match what you already know about the account. Once the analysis consistently holds up, move to supervised creation, then to launches you approve step by step.
Run One Command, Then Ask One Question
The connector turns Claude from a chat window you feed with exports into a working media-buying assistant that sits on live Meta data. The setup is a single command: claude mcp add --transport http adstellar https://app.adstellar.ai/api/mcp.
Once it shows as connected, begin with one read-only analysis prompt on your own account, such as asking which active ads are fatiguing and why. If the answer matches what you see in Ads Manager, you have a reliable foundation to build on. If it surfaces something you missed, you have already saved the time of the next export.
