Connect Meta Ads to Claude: the official Ads MCP, vendor connectors, or a synced backend

The most direct way to connect Meta Ads to Claude is Meta's own ads MCP server at https://mcp.facebook.com/ads. Add that URL in Claude as a custom connector — or in Claude Code as a remote HTTP server — sign in with Facebook, and Claude can pull campaign reports and create or edit campaigns, ad sets and ads. Meta announced these "Meta ads AI connectors" in open beta on April 29, 2026. Two other routes exist. Hosted vendor connectors such as Windsor.ai, Supermetrics and Pipeboard put Meta next to other ad platforms in one connector, and they can make changes too. A synced data backend such as TableBI stores campaign-level Meta data that your agent queries with read-only SQL, beside Google Ads, GA4 and Search Console. Pick Meta's server to let Claude operate a Meta account, a vendor to cover several networks from one connector, and a synced backend when the same report comes back every week and must never touch a live campaign.

Three ways to connect Meta Ads to Claude, compared

"Meta Ads MCP" can mean Meta's own server or someone else's. Meta hosts the official one. Vendors run their own hosted MCP servers on top of Meta's Marketing API, and all three covered here can change campaigns. A synced backend — the route the second half of this page sets up — only reads.

Sources, read September 26, 2026: Meta's announcement (April 29, 2026, updated July 16), Business Help Center article, developer docs and developer blog; Anthropic's custom connectors article; Windsor.ai's Meta Ads to Claude guide and pricing; Supermetrics' Facebook Ads to Claude page; pipeboard-co/meta-ads-mcp and Pipeboard pricing.

Pick by the job. If Claude should operate the account — draft campaigns, shift budget, pause what's fatigued — use a connector that writes, and set its guardrails before the first prompt: Meta's ads MCP server rules let a business-portfolio admin block campaign creation or cap budgets per ad account. Meta's own server is the direct route when Meta is the only platform in the conversation; a vendor connector makes sense when the same conversation also has to act on Google Ads or other networks. If the job is a number you report every week — spend, CPA and ROAS against Google Ads and GA4 under one definition — a synced backend fits better: the data is stored once, the agent can only read it, and a pinned dashboard refreshes without anyone re-asking the API.

Meta's official ads MCP server: what it does and how to add it

Meta ads AI connectors include a Meta-hosted ads MCP server and an ads command-line interface. Meta announced both in open beta on April 29, 2026. On July 16, 2026, it opened the server to developers with their own Meta app and added ads MCP server rules, which let anyone with full control of a business portfolio decide what AI agents may do on each ad account or catalog. Per Meta's Help Center, the server covers:

• Reporting — performance insights and detailed reports; Meta's developer blog lists spend, impressions, CTR and ROAS with breakdowns by age, gender and platform.
• Ad creation and management — create and edit campaigns, ad sets, ads and creatives. All ads are paused by default until you set them live.
• Audiences, catalogs and tests — create, update and delete custom audiences; build catalogs and troubleshoot feeds; create and manage A/B tests and conversion lift studies.
• Diagnostics — signal health for your pixel and datasets, a search of Meta Business Help Center articles, and activity logs that mirror the campaign history page in Ads Manager.

Meta lists four supported agents — ChatGPT, Claude, Claude Code and Perplexity — and the server URL is the same for all of them:

# Claude app: Customize → Connectors → + → Add custom connector
#   URL: https://mcp.facebook.com/ads

# Claude Code: add it as a remote HTTP server
claude mcp add --transport http meta-ads https://mcp.facebook.com/ads
# then run /mcp inside Claude Code and finish the Facebook login in the browser

In the Claude app, custom connectors work on every plan, but the Free plan allows one, and on Team and Enterprise plans an Owner adds the connector for the organization first. Developers who want the server under their own Meta app pass that app's ID with --client-id. Two cautions from Meta's own pages: access is still rolling out — the Help Center warns you may not have every tool yet — so ask the agent which tools it has after connecting; and Meta says any action taken on your behalf needs your authorization through the agent, so set the rules in Meta Business Suite (Settings → Integrations → Ads MCP server) before a budget is in play. For terminal-first work, Meta's ads CLI (meta ads …, Python 3.12 or later) exposes the same kind of operations as shell commands, and it also creates new resources paused by default.

Vendor connectors: one connector across ad networks

Windsor.ai, Supermetrics and Pipeboard each run a hosted MCP server that reaches Meta through its Marketing API and also covers other ad platforms, so one conversation can compare — or change — Meta and Google Ads together. All three write, and all three put a check in front of changes: Windsor.ai shows the change and waits for your confirmation, Supermetrics creates new campaigns paused and logs every action in its Campaign History, and Pipeboard confirms writes and starts new campaigns paused. Windsor.ai and Supermetrics are listed in Claude's connector directory; Pipeboard's README has you add its server by URL. Their prices scale differently: Windsor.ai's plans with data sources and accounts, Supermetrics' with data sources, users and accounts after a 14-day trial, and Pipeboard's with ad account slots and weekly tool executions.

The read-only route: sync Meta Ads into SQL for Claude Code

The third route doesn't give Claude the keys to the account. TableBI syncs Meta Ads into a data backend that Claude Code queries with read-only SQL, next to Google Ads, GA4 and Search Console:

npm i -g @tablebi/cli
tablebi login
tablebi install

# browser opens for Meta OAuth; the ad account starts syncing
tablebi connect meta_ads

TableBI is not an MCP server: tablebi install plants a Claude Code skill, and the agent runs tablebi ask for read-only SQL and tablebi pin for a live dashboard URL. An MCP entrance is on the roadmap, not shipped. The connector requests only Meta's ads_read permission, so nothing here can edit a campaign. Synced data lands at two altitudes:

• facts — one row per day per campaign: cost (Meta's spend), impressions, clicks, conversions and conversion_value, under the same definitions as Google Ads and GA4. The live connector syncs at campaign level, so date, platform, account and campaign carry values while ad_group, ad, channel and country stay empty.
• meta_ads_raw — Meta's own fields for the same rows, including reach, frequency, CPC and CPM, for when the unified view isn't enough.

One definition to know before the first query: the synced conversions count Meta's purchase actions, and conversion_value is their purchase value. A lead-generation campaign therefore shows zero conversions here — read that zero as "no purchases", not as waste.

Ad set and ad detail come in by file. Export an Ads Manager report broken down by day that includes the Ad set name column, and load it; the ad set lands in ad_group. When the same account also syncs live, give the export its own label, as below, so its rows don't add to the synced campaign totals. If the beta connector can't authorize your account, label the export meta_ads instead, so it sits exactly where synced rows would — the Facebook Ads to Looker Studio guide covers that export, including the Platform breakdown that splits out Instagram.

# ad set rows from an Ads Manager export, kept apart from the campaign sync
tablebi connect csv --file meta-adsets.csv --platform meta_ads_adsets

Three queries for the weekly Meta review

1. The campaign leaderboard: spend, purchases, CPA, ROAS

tablebi ask "SELECT campaign, SUM(cost) AS spend, SUM(conversions) AS purchases,
             cpa(SUM(cost), SUM(conversions)) AS cpa,
             roas(SUM(conversion_value), SUM(cost)) AS roas
             FROM facts
             WHERE platform = 'meta_ads'
               AND date >= (SELECT MAX(date) FROM facts WHERE platform = 'meta_ads') - 28
             GROUP BY campaign ORDER BY spend DESC LIMIT 20"

cpa(…) and roas(…) are definition macros: the arithmetic lives in the engine, so the agent never re-derives a ratio slightly wrong. The window anchors to Meta's own newest synced day rather than the wall clock — current_date isn't available in the query layer — so days that haven't synced can't drag the totals down.

2. Campaigns spending with zero purchases

tablebi ask "SELECT campaign, SUM(cost) AS spend, SUM(clicks) AS clicks
             FROM facts
             WHERE platform = 'meta_ads'
               AND date >= (SELECT MAX(date) FROM facts WHERE platform = 'meta_ads') - 28
             GROUP BY campaign
             HAVING SUM(cost) > 0 AND SUM(conversions) = 0
             ORDER BY spend DESC"

This is the scan that is tedious in Ads Manager and one pass here. Before acting on anything it lists, check two things: whether the campaign optimizes for something other than purchases, such as leads or traffic, and whether most of its spend falls in the last few days — Meta keeps crediting conversions within its attribution window, and the trust block warns that the most recent 72 hours move the most. The decision to pause stays with a person in Ads Manager — or with one of the connectors above, if you've chosen to let Claude write.

3. Meta versus Google Ads, one definition of ROAS

tablebi ask "SELECT platform, SUM(cost) AS spend,
             roas(SUM(conversion_value), SUM(cost)) AS roas,
             cpa(SUM(cost), SUM(conversions)) AS cpa
             FROM facts
             WHERE platform IN ('meta_ads', 'google_ads')
               AND date >= (SELECT MAX(date) FROM facts) - 28
             GROUP BY platform ORDER BY spend DESC"

Here platform is just another dimension, and cost and conversion value mean the same thing on every row, so a cross-network comparison is a GROUP BY. Treat the result as directional — Meta and Google attribute conversions differently, and when both are connected the trust block that ships with every answer attaches a caveat saying so, next to each source's freshness, so a two-row table can't hide it. The Google Ads side of this setup is covered in connecting Google Ads to Claude Code, and GA4 in connecting Google Analytics 4 to Claude.

Pin it: a Meta Ads weekly that refreshes itself

tablebi pin --title "Meta Ads weekly" \
  --widget "Daily spend (28d)::line=SELECT date, SUM(cost) AS spend
    FROM facts WHERE platform = 'meta_ads'
    AND date >= (SELECT MAX(date) FROM facts WHERE platform = 'meta_ads') - 28
    GROUP BY date ORDER BY date" \
  --widget "Campaign leaderboard=SELECT campaign, SUM(cost) AS spend,
    SUM(conversions) AS purchases, roas(SUM(conversion_value), SUM(cost)) AS roas
    FROM facts WHERE platform = 'meta_ads'
    AND date >= (SELECT MAX(date) FROM facts WHERE platform = 'meta_ads') - 28
    GROUP BY campaign ORDER BY spend DESC LIMIT 10"
✓ published → https://dk.tablebi.com/d/dsh_…  (public, read-only, self-refreshing)

Widgets tagged ::line render as line charts; the rest render as tables. The URL is public, read-only and self-refreshing — syncs run on a schedule and are also triggered when someone opens the page. The full reporting routine this replaces, from ROAS rankings to week-over-week checks, is in Meta Ads reporting. The same three-way comparison for the other sources: Google Ads, Google Analytics 4 and Search Console.

FAQ

npm i -g @tablebi/cli && tablebi install