Back to Guides Catalog
AI & MCP
4 min read
August 2026

The Developer & Agent Guide to MCP-Native Display Advertising

Model Context Protocol (MCP) lets autonomous AI agents buy campaigns and lets app developers provision ad placements programmatically — with hard money guardrails designed in.

Why Model Context Protocol (MCP) Changes Display Ads

Traditional ad platforms assume a human clicking through a dashboard, or an engineering team maintaining fragile REST integrations. Model Context Protocol changes the interface: an LLM-based agent connects to an MCP server and discovers typed, documented tools it can call directly — no scraping, no brittle API glue.

Birtingur runs a native MCP server at mcp.birtingur.app. Through it, AI agents (Claude, custom agentic pipelines, or anything else that speaks MCP) can inspect ad categories, check budgets, launch campaigns, and provision publisher ad slots. Advertising becomes infrastructure that agents can operate — under guardrails a human configured first.

What the Birtingur MCP Server Exposes

Every MCP session is scoped by its API key. Publisher-scoped keys see publisher tools: register_publisher, create_slot, update_slot, list_slots, get_snippet, get_react_component, get_stats, set_content_policy, and get_changelog. Advertiser-scoped keys see buying tools: list_categories, create_campaign, get_campaign, list_campaigns, list_creatives, and get_wallet.

The prerequisite is deliberate and non-negotiable: every MCP call requires an API key that a human created in the Birtingur dashboard. There is no self-serve agent onboarding and no zero-touch path from “agent exists” to “agent spends money” — a design choice, not a missing feature.

For Advertisers: Programmatic AI Agent Purchasing

Using tools like list_categories and create_campaign, an agent can evaluate campaign objectives, pick relevant content verticals such as Tech & Innovation, set a budget, and submit the buy — all at the same flat 550 kr. CPM a human buyer pays.

Spending is fenced in three ways. Purchasing must be explicitly enabled per API key by its owner. Each key’s spending counts against a monthly cap. And any single purchase above the key’s auto-approve limit is created in a pending state and waits for the owner to approve or reject it in the dashboard — the agent cannot approve its own purchase. Campaign creation also supports idempotency keys, so an agent retrying a request cannot accidentally buy twice.

For Developers: Ad Infrastructure Provisioning

With a publisher API key created in the dashboard, developers building web apps or AI tools can call register_publisher, create_slot, and get_react_component via MCP. The result is a fully typed React component — a BirtingurAdSlot — that reserves its dimensions to avoid layout shift, handles loading and fallback states, and fires viewability-correct impression tracking.

For non-React sites, get_snippet returns the standard embed instead: a single script under 5KB. Either way the integration serves privacy-first contextual ads with zero cookies, and revenue is shared 80/20 with monthly payouts. Our AI app monetization guide covers the developer path end to end.

Guardrails Are the Feature

Agentic buying only works commercially if money cannot leak. That is why the boundaries sit server-side, keyed to the API key’s identity rather than to anything the agent claims about itself: wallet top-ups, budget increases, and refunds are dashboard-only operations with no MCP path at all, and high-value purchases always pause for a human.

For teams building agents, this is the pitch: your agent gets real buying power inside limits you set, and the failure mode of a confused agent is a pending approval in your dashboard — not a drained wallet. That trade is what makes it reasonable to hand advertising to software in the first place.

Frequently Asked Questions

Can an AI agent spend money without human oversight?

No. Purchasing must be enabled per API key by its human owner, spending counts against a monthly cap per key, and purchases above the key’s auto-approve limit wait for human approval in the dashboard. Top-ups and refunds have no MCP path at all.

Where is the MCP server and how do I authenticate?

The server runs at mcp.birtingur.app. Authentication is a bearer API key created by a human in the Birtingur dashboard; the key’s scope (publisher, advertiser, or both) decides which tools the session sees.

What React support exists?

The get_react_component tool returns a typed BirtingurAdSlot component that reserves its dimensions (no layout shift), handles loading and fallbacks, and implements viewability-correct impression tracking. Non-React sites use get_snippet for the standard sub-5KB embed.

Can an agent register a publisher account from scratch?

An agent can call register_publisher — but only with an API key a human already created in the dashboard. There is deliberately no zero-touch onboarding path for agents.

What happens if my agent retries a purchase request?

Campaign creation supports idempotency keys, so a retried request with the same key resolves to the same campaign instead of creating a duplicate purchase.

Ready for Privacy-First Display Ads?

Join creators and brands testing Birtingur’s cookie-free category display network.

Join Early Access Waitlist