Claude for Meta Ads: Connector, Skills and 25 Prompts

Details below are as of October 2026. Meta's server is in open beta, so tool names and permissions can move.

Anthropic draws the line in one sentence: MCP gives Claude access to tools, skills teach Claude how to use them. The third word comes from elsewhere.

Why a skill alone does nothing

A skill is knowledge without access. A "Meta Ads audit skill" can tell Claude to check frequency, learning status and Event Match Quality, but it still needs the connector (or an export you upload) to have numbers to check. Skills are available on Free, Pro, Max, Team and Enterprise, need code execution switched on, are enabled under Customize, then Skills, and follow an open standard published at agentskills.io. The same skill also works in Claude Code and through the API.

Google has gone further than Meta here: it publishes ads skills for its own API and a Google Ads API Developer Assistant plugin for Claude Code. See Claude for Google Ads for that side.

Meta has no ads connector in Claude's directory, so you add Meta's server yourself. The full walkthrough, including agencies and Advanced Access, is in Meta Ads MCP: setting up Meta's official server. The short version:

1. In Claude, open Connectors in settings and choose Add custom connector.
2. Paste https://mcp.facebook.com/ads, name it Meta Ads, and connect.
3. Sign in with Facebook Login for Business, pick the business portfolios, approve the permissions.
4. Start with ads_mcp_management and ads_read. Add ads_management only when you want Claude to create drafts.
5. Test with a read-only prompt: "List my ad accounts and the last 7 days' spend in each."

Plan limits to know: custom connectors work on every Claude plan, but Free allows one, which matters as soon as you want a second connector for revenue next to Meta. On Team and Enterprise an Owner adds the connector for the organisation and each member connects their own login. In Claude Code, Meta documents the command:

claude mcp add --transport http --client-id <META_APP_ID> meta-ads https://mcp.facebook.com/ads

Most teams run the same three or four analyses every week. Writing them down once as a skill saves retyping the context and makes the answers comparable from week to week. What the instructions should pin down:

• Which accounts and which goalThe ad accounts in scope and the one result each campaign is judged on, so Claude never compares a lead campaign with a traffic campaign on CPC.
• Which periodsLast 7 days against the 7 before, or month to date against last month. A fixed comparison is what makes two weeks of reports readable side by side.
• What to check, in orderSpend pacing, cost per result, frequency, anomalies, dataset quality, recent changes from the activity log. Your checklist, written out.
• How to present itOne table per account, the three findings that need a decision first, and a line saying which tool and date range each number came from.
• What it may not doNever activate anything, never change a live budget, always propose changes as a table for a person to apply.

Keep the skill about method and leave the account data to the connector. A skill that hardcodes last month's numbers goes stale the day you save it.

Reporting runs mostly on ads_get_ad_entities, which returns spend, impressions, CTR, CPC, CPM and conversions with filters, breakdowns and date ranges, plus the trend and benchmark tools. Name the account and the period in every prompt.

1. "For the [account] ad account, show spend, results and cost per result per campaign for the last 7 days against the 7 days before, as a table sorted by spend."
2. "Which ad sets spent more than [amount] in the last 14 days with no conversions? List them with spend, impressions and frequency."
3. "Break last month's spend and conversions down by placement and by age. Where is the cost per result more than 50% above the account average?"
4. "Show the performance trend for cost per result and CTR over the last 30 days for our three biggest campaigns, and tell me on which days it moved most."
5. "Are there anomaly signals on this account this week? For each, name the campaign, the metric and what changed."
6. "Compare our CPM and CTR with the industry and auction ranking benchmarks Meta has for this account. Where are we below?"
7. "Write a five line weekly summary for the [account] account: spend, results, cost per result, the best and worst campaign, and one thing to check."

Audits combine the reporting tools with the opportunity score (0 to 100 with recommendations), the dataset tools and the activity log.

1. "What is this account's opportunity score, and which of Meta's recommendations would change delivery rather than just tidy up settings?"
2. "Check the datasets on this account: Event Match Quality, match key coverage and freshness for the Lead event. What is missing?"
3. "Show event volume per day for our main conversion event over the last 28 days. Are there days where it drops to near zero?"
4. "List every change made to campaigns and ad sets in the last 14 days from the activity log, and put it next to the daily cost per result."
5. "Find ad sets whose audiences overlap in targeting, using the custom audiences and lookalikes attached to each."
6. "Which campaigns optimise for an event other than the one we report on? List the optimisation goal of each ad set."

Prompt 9 is the one to run first on a lead generation account. If Event Match Quality is weak, Meta is optimising on a blurry signal, and every other audit finding is downstream of that. Event Match Quality for lead generation explains what moves it.

Creative review uses ads_get_creatives, ads_get_creative_ads, ads_get_ad_preview, the image and video listing tools and ads_library_search for the public Ad Library. Meta does not document that the server returns video files or transcripts, so these prompts work on copy, headlines, calls to action and results.

1. "List the active ads in [campaign] with their primary text, headline and call to action, plus spend and cost per result for the last 30 days."
2. "Group our active ads by the promise in their headline (price, speed, outcome, social proof) and show cost per result per group."
3. "Which creatives run in more than one ad set, and how does cost per result differ between them?"
4. "Search the Ad Library for active ads from [competitor] and summarise the offers and calls to action they repeat most."
5. "Show me the preview of our three ads with the highest frequency and lowest CTR in the last 14 days. Which look worn out?"
6. "Draft three new primary text variants for our best ad by cost per result, keeping the offer and changing only the opening line."

Read prompt 15 with care: cost per result ranks creatives by how cheaply they produce form fills, which is not the same as how well they produce customers. More on that below.

With ads_management approved, Claude can create campaigns, ad sets, creatives and ads. Meta's server creates all of them paused; spend starts only through ads_activate_entity. No dry run is documented, and there is no tool to delete a campaign, ad set or ad, so drafts you do not want stay in the account until you remove them in Ads Manager.

1. "Create a paused lead generation campaign called [name] with one ad set: [country], ages [range], daily budget [amount], optimising for leads. Do not activate it."
2. "Duplicate the structure of [campaign] as a new paused campaign for [new country], with the same ad set settings and budgets halved."
3. "Create a single image link ad in the paused ad set [name] using the image [name] from the account's image library, this headline and this primary text. Show me the preview."
4. "Build a lookalike audience from our customer list audience [name] and attach it to a new paused ad set in [campaign]."
5. "Before changing anything: propose new daily budgets for the ad sets in [campaign] based on last 14 days' cost per result, as a table of current and proposed values."
6. "List everything you created in this conversation, with IDs and status, so I can review it in Ads Manager."

Every number in the 25 prompts comes from Meta: what it delivered, what your pixel and Conversions API reported, and the values you sent. For lead generation that ends at the form. Meta does not know which leads your sales team qualified, which booked a call or which signed, so when you ask Claude for the best campaign, it ranks by cost per lead.

Meta's data says move budget from B to A. The CRM says the opposite. Claude is not wrong in the first answer; it was given half the data. Why your ad platform's MCP gives Claude the wrong CPL goes deeper, and sending the CRM stages back to Meta (see send CRM data back to Meta and Google) fixes what Meta optimises on, but not what Claude can read through Meta's server today.

LeadJourney tracks every Meta click, form fill and call server-side and first-party, keeps the fbclid on the lead, and follows it into HubSpot, Salesforce, Pipedrive, Close, Attio, GoHighLevel, ActiveCampaign, Odoo, Zoho CRM or Microsoft Dynamics 365. So each Meta campaign, ad set and ad carries its qualified leads, booked demos and closed revenue next to the spend. Setup takes about 21 minutes, with 95%+ tracking accuracy.

Its MCP server is the second connector in the same Claude conversation: https://api.leadjourney.io/mcp, signed in with OAuth (setup for Claude). It reads journeys, spend per campaign and ad set, and CRM outcomes, and it changes only things inside LeadJourney, such as reports and the attribution model. It cannot pause a campaign, move budget or create an ad. With both connected, prompts 1, 2, 15 and 24 can be asked on closed revenue instead of cost per lead:

• "From LeadJourney, closed revenue and cost per qualified lead per Meta campaign last quarter. From Meta, current daily budgets. Which three campaigns are underfunded for what they close?"
• "Rank our active Meta ads by closed revenue in LeadJourney rather than by cost per result, and show where the two rankings disagree."

The MCP connection is included from LeadJourney's Growth plan and is a €49 a month add-on below it (€99 a month flat for agencies). The full two-connector workflow, with guardrails, is in Meta Ads MCP plus your CRM.

• Meta Ads MCP: setting up Meta's official server
• Meta Ads MCP plus your CRM: which campaigns made revenue
• Ads MCP servers compared: Meta, Google, LinkedIn, TikTok and more
• Claude for Google Ads: connect it, then ask better questions
• Why your ad platform's MCP gives Claude the wrong CPL
• AI ad creative analysis: hook, body and close, judged on revenue
• Which ad copy attracts leads that close
• LeadJourney in Claude
• The Meta Ads integration