AI advertising glossary
MCP server for ads
An MCP server for ads exposes ad-platform actions as Model Context Protocol tools, so an AI agent can plan, fund and report on campaigns.
What it means
The Model Context Protocol (MCP) is a standard way for AI assistants to discover and call tools. An ads MCP server turns platform actions such as listing channels, creating a draft or reading a report into tools an agent can call without custom integration code.
Admux runs an MCP server over Streamable HTTP at /mcp on its API host, authenticated with the same Bearer key as the HTTP API, and a stdio version for local use. Its tools mirror the API operations: channels_list, asset_upload_image, campaign_create_draft, campaign_validate, offer_create, campaign_submit, campaign_launch, campaign_pause, balance_get, report_get and more. Tools the key lacks scopes for are not listed, so an agent sees only what it may do.
Gravity also has a dashboard MCP server. It uses the same single API key as Gravity’s REST API, has full admin access, runs actions without confirmation and has no spend limit.
When you buy ads
Connect an agent through MCP when you want it to work inside a chat or coding assistant, and use the HTTP API for scripts and back-office jobs; in Admux both enforce the same scopes. Give each agent its own key with a spending limit, and leave launch scope off until you trust its drafts. Tools that move money or change delivery need an idempotency key, so a retried call does not submit a campaign twice. Treat a submit or pause call as a request and have the agent poll the operation status before reporting success.
Sources
- Dashboard MCP server for advertisers - Gravity docsGravity docs · checked 10 Nov 2026
- Advertiser APIs - Gravity docsGravity docs · checked 10 Nov 2026