The Publisher's Guide to the Prebid Sales Agent

Everything publishers need to know about the open-source Prebid Sales Agent and the emerging agentic future of media sales.

The organization that set the standard for header bidding has introduced something new. In January 2026, Prebid.org announced the Prebid Sales Agent, an open-source, AI-enabled seller agent that gives publishers direct access to agentic advertising demand.

Built on the Ad Context Protocol (ADCP), it allows AI buyer agents from platforms such as Claude, ChatGPT, and Google Gemini to discover publisher inventory, negotiate terms, and complete media buys automatically at machine speed.

For publishers who have heard about ADCP and agentic advertising but are unsure how to begin, this guide explains what the Prebid Sales Agent is, how it works, and how it connects to your existing publisher revenue strategy. For a broader introduction to ADCP itself, see our post on BiddingStack and agentic advertising.

Table of Contents

1. What's the Prebid Sales Agent?
2. Why Prebid?
3. How the Prebid Sales Agent Works
4. The ADCP Media Buy Protocol
5. Benefits for Publishers
6. Getting Started: Deployment Options
7. Prebid Sales Agent vs. Managed Seller Agent Services
8. The System Growing Around Prebid Sales Agent
9. BiddingStack and Prebid Sales Agent
10. Current State of the Prebid Sales Agent

What Is the Prebid Sales Agent?

The Prebid Sales Agent is an open-source reference implementation of a publisher-side AI agent that automates programmatic ad sales. Published on GitHub at github.com/prebid/salesagent, it's freely available to any publisher, ad tech platform, or developer without requiring Prebid membership or a commercial license.

The Sales Agent performs the same role as a traditional human ad sales team, but it operates autonomously and at scale. It exposes your inventory to AI buyer agents in a machine-readable format, responds to campaign briefs written in natural language, negotiates deal terms (including targeting and placement), executes transactions end-to-end without human intervention, and manages the entire campaign lifecycle from discovery through reporting.

Think of it as giving your ad operations team an AI-powered colleague that never sleeps, can handle thousands of simultaneous buyer inquiries, and speaks fluent ADCP.

This isn't another layer of automation on top of existing programmatic plumbing. The Prebid Sales Agent represents a different channel in which AI buyer agents initiate and complete media purchases through structured, conversational agent-to-agent communication rather than relying on real-time auction signals.

Why Prebid?

When header bidding emerged as an alternative to the Google Ad Manager waterfall, it was fragmented and inconsistent across implementations. Prebid created an open, vendor-neutral standard, and the industry quickly rallied around it. Today, Prebid.js powers header bidding on hundreds of thousands of publisher websites globally, with support for over 300 demand partners.

Prebid is applying that same playbook to agentic advertising. After the Ad Context Protocol was incubated within the Agentic Advertising Organization (AAO), Prebid took stewardship of the sell-side agent codebase.

The governance split is intentional and important. Prebid leads software development and maintains the Sales Agent repository, while AgenticAdvertising.org governs the ADCP protocol specification and compliance.

This separation ensures the software layer evolves through community contribution while the protocol standard remains neutral territory. Publishers get the same guarantees that made Prebid.js trustworthy, including no vendor lock-in, transparent development, and a community with aligned interests.

How the Prebid Sales Agent Works

The Sales Agent is built in Python and uses two industry-standard agent communication protocols to expose inventory to AI buyers.

MCP (Model Context Protocol)

MCP is the emerging standard for AI systems to expose capabilities and data to AI agents. The Prebid Sales Agent uses FastMCP with HTTP/SSE transport to present your inventory as a set of structured tools and resources. AI agents, including Claude, ChatGPT plugins and Google Gemini, can call these tools directly. When a buyer's AI assistant queries your Sales Agent via MCP, it receives structured information about what you sell, your audience segments, available formats, and pricing.

A2A (Agent-to-Agent) Protocol

For agent-to-agent transactions that don't route through a specific AI assistant, the Sales Agent implements a JSON-RPC 2.0-compliant server at the /a2a endpoint. Discovery happens through a standard /.well-known/agent.json manifest, making your Sales Agent automatically discoverable by buyer agents that scan for available seller endpoints.

Both protocols are served from a single deployment. Buyers access whichever interface suits their architecture.

The Admin Interface

Publishers manage their Sales Agent through a web-based Admin UI, accessible at your deployment URL. From this interface, you can configure your ad server integration, with Google Ad Manager supported out of the box. You can also define products that map to your GAM line items and set pricing floors and deal rules. Before execution, deals can be reviewed and approved in human-in-the-loop mode. You can also monitor campaign performance and transaction logs.

Every transaction is a logged, verifiable conversation between buyer and seller agents, creating an audit trail that neither party can dispute.

The ADCP Media Buy Protocol

The Prebid Sales Agent implements the Ad Context Protocol (ADCP) Media Buy protocol, which defines 10 standardized task types covering the complete advertising lifecycle:

Task	Description
Discovery	AI agents find relevant inventory using natural language briefs
Proposal	Agent generates deal proposals matching buyer requirements
Negotiation	Back-and-forth on pricing, targeting, and terms
Approval	Human-in-the-loop or automated sign-off
Activation	Campaign goes live in your ad server
Pacing	Delivery management against contracted terms
Optimization	Real-time adjustments to improve performance
Reporting	Standardized performance data returned to buyer agent
Billing	Automated reconciliation and invoicing support
Attribution	Outcome measurement and verification

This protocol sits above the auction layer. Rather than competing at the impression level, buyer agents engage with your Sales Agent to structure larger, pre-negotiated deals. This approach reduces auction overhead and helps build more strategic, high-value relationships.

Benefits for Publishers

Access Agentic Demand Without Building AI Infrastructure

Most publishers are interested in agentic advertising, but the real challenge lies in the technical work needed to implement it. Building and maintaining an AI agent that speaks ADCP, integrates with your ad server, and stays current with protocol updates would require a dedicated engineering team. The Prebid Sales Agent eliminates this barrier entirely, which means download, configure, and deploy in under an hour.

No Vendor Lock-In

Because the Sales Agent is open source and built on open protocols, you're not committed to any single vendor's AI stack or commercial terms. Modify the reference implementation to fit your needs, contribute improvements back to the community, or engage service providers for managed hosting. It is your choice, your data, and your relationships.

Sell Non-Standard Inventory at Scale

The programmatic world does well at standard IAB display and video units. Custom sponsorships, newsletter placements, podcast integrations, AI chat placements, and branded content remain difficult to sell programmatically because OpenRTB wasn't designed for them. ADCP and the Prebid Sales Agent handle these formats natively, buyer agents can understand and transact any inventory type you can describe in natural language.

Expand Your Demand Footprint

When your Sales Agent is live and discoverable, buyer agents operating globally can find and transact with you without an existing commercial relationship. You don't need an insertion order or a meeting. Your agent handles demand partner discovery and qualification, surfacing only the deal opportunities that meet your minimum terms.

Maintain Control Over Business Rules

The Sales Agent works as your personal agent. You set pricing floors, blacklist categories, approve deal structures, and define the rules of engagement. Unlike open-auction programmatic, where visibility into buyers and their intentions is limited, every agentic transaction is explicit, documented, and fully governed by your rules.

Reduce Manual Sales Operations Overhead

For publishers with direct sales teams, the Sales Agent handles inbound inquiries, generates proposals, and manages routine AI-powered ad sales operations. Your human team can focus on strategy and high-value custom packages, which is the work that actually requires human judgment.

Getting Started: Deployment Options

Prebid offers three deployment paths depending on your infrastructure preferences:

Option 1: Docker (Local / On-Premises) ~2 minutes

The fastest path to a running Sales Agent. Docker Compose brings up the full stack, including PostgreSQL, with a single command. Ideal for testing, development, or publishers with existing on-premises infrastructure.

git clone https://github.com/prebid/salesagent
cd salesagent
docker compose up

Access the Admin UI at http://localhost:8000 and begin configuration.

Option 2: Fly.io (Cloud) ~15 minutes

For publishers who prefer a managed cloud deployment without maintaining servers, Fly.io offers a straightforward path with minimal infrastructure overhead.

Option 3: Google Cloud Run (Cloud) ~20 minutes

For publishers already invested in the Google Cloud world, Cloud Run provides a serverless, expandable deployment that integrates naturally with Google Ad Manager and existing GCP tooling.

First Configuration Steps

Regardless of deployment method, your initial configuration follows the same path:

1. Open the Admin UI and go to Settings > Adapters
2. Connect your Google Ad Manager account
3. Create Products that correspond to your GAM line items and inventory offerings
4. Define pricing, targeting parameters, and deal rules for each product
5. Test your agent endpoint using an ADCP-compatible buyer agent or the Prebid testing tools
6. Submit your agent discovery URL to relevant buyer agent registries

No Prebid membership is required. The repository is open for direct contribution and modification.

Prebid Sales Agent vs. Managed Seller Agent Services

The Prebid Sales Agent is a reference implementation, providing a complete, functional starting point. Publishers who opt for the self-hosted route are responsible for ongoing maintenance as the ADCP protocol evolves (currently moving toward v3), ensuring infrastructure reliability, scaling, and uptime, keeping ad server adapters updated and compatible, managing security patches and dependencies, and handling monitoring, alerting, and incident response.

For publishers with engineering resources and a preference for full control, self-hosting is a legitimate choice. For publishers who want to capture agentic demand without the operational overhead, managed Seller Agent services, like those offered by BiddingStack, provide the same capabilities without the infrastructure burden.

The distinction mirrors the choice between self-hosting Prebid Server and using a managed Prebid Server provider. The technology is the same, the question is who manages it.

The System Growing Around Prebid Sales Agent

The Prebid Sales Agent launched with meaningful industry backing. Major SSPs and platforms, including Magnite and PubMatic, helped define how Seller Agents should function. Early ADCP coalition members span the full supply chain. They include Scope3 for emissions and quality measurement, Triton Digital for audio and streaming inventory, PubMatic and Magnite for SSP infrastructure, Optable for privacy-preserving audience data, and Yahoo for publisher and demand network support.

As buyer-side AI agents multiply, and as platforms like Claude, ChatGPT, and Gemini deepen their advertising integrations, the volume of agentic buy-side demand will grow rapidly. Publishers who have a live Sales Agent endpoint will capture this demand. Publishers without one will not.

The window for early-mover advantage in agentic advertising is open now, but it will not remain open indefinitely.

BiddingStack and Prebid Sales Agent

BiddingStack is an ADCP Founding Member and an active participant in the Prebid Sales Agent environment. Our ADCP Seller Agent services build on the Prebid Sales Agent foundation to deliver:

• Managed hosting and operations: we run the infrastructure so you don't have to
• Protocol updates: as ADCP evolves toward v3 and beyond, your agent stays current automatically
• Unified header bidding integration: your agentic demand and programmatic demand are optimized together, not in silos
• Floor price automation: AI-driven pricing logic applies across both agentic and RTB demand, so every deal meets your revenue targets
• Non-standard inventory support: AI chat, recommendation, search, and conversational placements alongside standard display and video
• Publisher network scale: for networks managing multiple publisher properties, our infrastructure handles multi-tenant deployments with centralized controls

Publishers using BiddingStack typically see yield improvements of 50% to 200% compared to similar vendors across their existing programmatic inventory. Adding agentic demand through ADCP creates an additional, incremental revenue channel on top of that foundation.

Whether you want to self-host the Prebid Sales Agent or build on BiddingStack's managed services, the path to agentic advertising starts with the same open standard.

Current State of the Prebid Sales Agent

The Prebid Sales Agent is a living project under active development. The ADCP protocol is advancing toward v3 with expanded content standards and enhanced targeting capabilities. Buyer-side agent development is accelerating across every major AI platform. The infrastructure is being built now, and the demand will follow.

For publishers, the strategic calculus is straightforward. The cost of early adoption is low and the cost of waiting is asymmetric. An operational Seller Agent endpoint costs relatively little to maintain, while the revenue opportunity from being discoverable by AI buyer agents grows as adoption scales.

Prebid is repeating its success from header bidding in agentic advertising. It creates a fair and open standard that levels the playing field and expands the opportunity. Publishers who rode the header bidding wave early captured sizable yield improvements. The agentic wave is beginning now.

Ready to Activate Your Seller Agent?

BiddingStack helps publishers deploy and operate ADCP-compliant Seller Agents without the infrastructure overhead. Whether you're evaluating the Prebid Sales Agent reference implementation or ready for a managed deployment, our team can help you get live quickly.

Contact us or visit BiddingStack.com to learn more about our ADCP Seller Agent services and how they complement your existing header bidding setup.