Broker — Agentic trading
✦ Coming SoonThe next evolution of programmatic advertising: AI agents that discover inventory, activate audience signals, execute campaigns, and optimise bids — all autonomously, all in real time.
What is Broker?
Broker is an agentic trading layer built on top of the exchange. Instead of manually configuring demand source (DSP) connections, floor prices, and campaign budgets through a UI, you describe your objective in natural language — and a chain of AI agents handles the execution.
“Activate premium CTV inventory in the Gulf for gaming advertisers at a $12 CPM floor.”
Broker translates that into live exchange operations: signal discovery, demand source targeting, real-time bid execution, and continuous optimisation.
The Protocol: AdCP
Broker is built on AdCP (Ad Context Protocol) — an open industry standard that enables AI agents to interact with advertising infrastructure through structured, machine-readable interfaces.
AdCP is built on two foundations:
| Protocol | Role |
|---|---|
| MCP (Model Context Protocol, Anthropic) | Direct AI-assistant integration — synchronous and asynchronous task execution |
| A2A (Agent-to-Agent) | Complex multi-agent workflows and collaborative reasoning |
AdCP defines a task-first architecture: every advertising operation is a discrete, composable task with a JSON schema. Agents can discover capabilities, chain tasks, and monitor progress asynchronously.
Three Core Protocols
1. Signals Activation
Discover and deploy audience and contextual signals to decisioning platforms.
get_signals → find segments matching natural language criteria
activate_signal → deploy to DSPs/SSPs, get platform-specific segment IDsExample: An agent queries “High-income millennials interested in gaming, Gulf region” — receives ranked signal options with CPM pricing and coverage across connected platforms, then activates on selected demand sources automatically.
Supported signal types:
- Audience (demographic, behavioural, purchase intent)
- Contextual (genre, content category, keyword)
- Geographic (country, city, postal)
- Temporal (daypart, seasonality)
2. Media Buy
Full campaign lifecycle management — from inventory discovery to real-time delivery reporting.
get_products → discover inventory with natural language
create_media_buy → launch campaigns across platforms
get_media_buy_delivery → live performance metricsSupports human-in-the-loop approval workflows: the agent surfaces options, a human approves, execution begins automatically.
3. Creative
AI-powered creative generation and asset management.
build_creative → generate creatives from a brief
preview_creative → render visual preview (batch mode: 5–10× faster)
list_creative_formats → discover format specificationsCreatives generated by Broker agents are automatically validated against Praevis platform policies before being submitted to the exchange.
How It Connects to the Exchange
Broker agents interact with the exchange through the same infrastructure that today serves demand sources and supply sources — but via the AdCP protocol layer rather than raw OpenRTB.
AI Agent
│
├── AdCP (MCP / A2A)
│ │
│ ├── Signals Activation ──► DSP targeting configs
│ ├── Media Buy ──► Praevis Exchange auction engine
│ └── Creative ──► VAST / banner delivery
│
└── AXE (Agentic eXecution Engine)
├── Brand safety enforcement
├── Frequency capping
├── First-party data activation
└── Dynamic audience targeting at impression timeAXE (Agentic eXecution Engine) is the real-time enforcement layer — it intercepts each impression, applies agent-defined rules, and connects to the exchange via key-value pairs or standard RTB.
Integration: adagents.json
Supply sources (Publishers) and platforms that support Broker declare their capabilities in a standard file — adagents.json — hosted at the root of their domain. This tells agent clients which AdCP protocols and tasks are supported, and what inventory is available.
{
"version": "2.5",
"publisher_id": "pub_abc123",
"protocols": ["signals", "media-buy", "creative"],
"tasks": ["get_signals", "activate_signal", "get_products", "create_media_buy"],
"authorized_properties": [
"gaming.example.com",
"sports.example.com"
]
}A hosted builder for adagents.json will be available at adcontextprotocol.org/adagents.
Why It Matters
| Today | With Broker |
|---|---|
| Configure demand source connections manually | Agent discovers and connects to demand sources automatically |
| Set floor prices by supply source | Agent optimises floors per segment in real time |
| Monthly budget reconciliation | Sub-second spend pacing across regions |
| 15+ custom platform integrations | One AdCP endpoint per platform type |
| Campaign setup: hours | Campaign setup: seconds |
For MENA-focused operators with heterogeneous platform stacks, Broker reduces integration overhead to a single protocol — then lets AI handle the operational complexity.
Governance
AdCP is governed by the Agentic Advertising Organisation (AAO) — a neutral non-profit trade association (forming), structured like Prebid. Open source under Apache 2.0. Working groups meet on the first Wednesday of each month.
Praevis is an early contributor to the AdCP specification and will ship Broker as a native exchange feature — not a separate product.
Timeline & Early Access
Broker is in private design preview. The AdCP specification is at RFC v0.1. We are implementing the Signals and Media Buy protocols first, targeting a limited partner pilot in Q3 2026.
Planned milestones:
| Quarter | Milestone |
|---|---|
| Q2 2026 | Internal AdCP server + Signals protocol integration |
| Q3 2026 | Media Buy protocol — limited partner pilot |
| Q4 2026 | Creative protocol + AXE enforcement layer |
| 2027 | General availability |
Join the early access list: [email protected]
Related
- Delegate — today’s co-located bidding engine (JSON strategy, sub-1ms)
- Pathfinder — smart DSP routing that Broker will extend
- AdCP Protocol — upstream open-source specification