Meta's own ads connectors, in open beta: a hosted MCP server that lets an AI assistant work in your ad account, and a command-line tool (Ads CLI) for scripts.
- Kind
- Connector
- Keeplight status
- Researched, not yet tested
- Cost
- As of 2026-11: no price on Meta's docs. You pay normal ad spend. Confirm the beta terms with Meta.
- Connects to
- Meta Marketing API, Facebook ads, Instagram ads, Product catalogs, Meta Pixel and datasets
- Official site
- developers.facebook.com/documentation/ads-commerce/ads-ai-connectors/ads-mcp-server/ads-mcp-server-overview
What it does
In 2026 Meta released two official ways to work with your ads from AI tools. They are separate products. The ads MCP server is hosted by Meta at mcp.facebook.com/ads and works with AI assistants that support MCP. Its tools cover reporting, creating and editing campaigns, catalogs, datasets, A/B tests, and the account activity log.
The Ads CLI is a command-line tool that wraps the Meta Marketing API. You run commands such as "meta ads campaign list" and sign in with a Meta system user access token. Meta's setup guide installs it with Homebrew or pip, on macOS or Linux.
The tools can write as well as read. Meta says the Ads CLI creates campaigns, ad sets, and ads in PAUSED status by default, so a person has to turn them on. Meta's MCP server pages do not state the same default, so check each launch. It suits a team that wants an assistant to read results and draft changes while a person approves them.
Use cases
Ask for a plain-language ad performance summary
Connect the MCP server to your AI assistant and ask which campaigns spent the most and which cost the most per result. The reporting tools read the numbers straight from the account.
Pull ad numbers into a script with the CLI
A command such as "meta ads insights get" returns spend, impressions, and click-through rate, so a script can save them to a sheet each week.
Draft a new campaign for a person to approve
The assistant builds the campaign, ad set, and ads. Through the Ads CLI they start paused, so someone reviews and activates them. Meta's MCP pages do not state a paused default, so check each launch.
Check that tracking is healthy
The dataset and signal tools report on data quality, which helps a team catch a broken pixel or missing events early.
How to start
- Decide which one you need. The MCP server is for AI assistants, and the Ads CLI is for terminals and scripts.
- For the MCP server, add https://mcp.facebook.com/ads as a connector in an AI tool that supports MCP, then follow Meta's Get started page to sign in If you connect your own app, you need a Meta developer app with the ads MCP server use case, and a system user token must have the Employee role (Admin tokens cannot reach it).
- For the CLI, follow the Ads CLI Get started page. It installs with Homebrew or "pip install meta-ads" (Python 3.12 or newer, macOS or Linux).
- For the CLI, create a system user in Business Settings, give it access to your ad account, and generate a token with the scopes Meta lists (business_management, ads_management, pages_show_list, pages_read_engagement, pages_manage_ads, read_insights).
- Save the token and ad account ID in a .env file, then run "meta auth status" to check the setup.
- Start with read-only commands such as "meta ads campaign list" before you let anything create or change ads.
Visit Meta ads AI connectors (MCP server and Ads CLI)
Researched, not yet tested: Keeplight has read the docs but has not run it yet. Prices and features change, so check the official site before you buy.