Pubmatic

Defining the Future of AI in Programmatic Advertising

This repository contains the official specifications for PubMatic's Model Context Protocol (MCP) server implementations, enabling agent-to-agent communication for programmatic advertising workflows.

Overview

PubMatic is among the first in the ad tech industry to publish a public specification for deal management using AI agent-to-agent communication protocols. This pioneering initiative aims to transform how participants in the programmatic ecosystem communicate and collaborate, moving beyond static APIs and manual processes to dynamic, context-aware, and autonomous communications between AI agents.

Vision

Programmatic supply chains work best when information flows freely and securely between partners. With agentic AI and emerging protocols like MCP (Model Context Protocol) and A2A (Agent2Agent Protocol), we can finally move beyond static, manual communication and into a world where agents collaborate directly to solve for common goals:

• Less friction
• More transparency
• Stronger outcomes for buyers and publishers alike

Available Specifications

This repository contains specifications for the following tools:

• Deal Management: Tools for deal creation and troubleshooting, enabling seamless communication between publishers, buyers, and DSPs
• Inventory Discovery: Buyer-focused tools to discover and curate supply packages with structured, machine-readable outputs

Benefits of Agent-to-Agent Communication

Current Workflow

Today, when a buyer sees an underdelivering PMP deal, they reach out to their DSP contact. The DSP investigates and emails PubMatic. After multiple log checks and message exchanges, the issue may be identified days later.

sequenceDiagram
    participant Buyer
    participant DSP as DSP Contact
    participant PubMatic
    
    Buyer->>DSP: Report underdelivering deal
    DSP->>DSP: Initial investigation
    DSP->>PubMatic: Email support request
    PubMatic->>PubMatic: Log analysis
    PubMatic-->>DSP: Email response (hours/days)
    DSP-->>Buyer: Relay information
    Note over Buyer,PubMatic: Process takes days

Future Workflow

With agent-to-agent communication, a buyer's AI agent can ask a DSP's AI agent why the deal isn't delivering. The DSP agent queries PubMatic's server and receives instant root cause analysis, such as pending creative approval or advertiser block, along with fix suggestions. The DSP agent relays the answer immediately, allowing buyers and publishers to take action and unblock deals in minutes instead of days.

sequenceDiagram
    participant Buyer
    participant BuyerAI as Buyer AI Assistant
    participant DSPAI as DSP AI Agent
    participant PubMCP as PubMatic MCP Server
    
    Buyer->>BuyerAI: "Why is deal XYZ123 underperforming?"
    BuyerAI->>DSPAI: Query deal status
    DSPAI->>PubMCP: API call to deal_troubleshooting
    PubMCP-->>DSPAI: Structured response with root cause
    DSPAI-->>BuyerAI: Relay analysis and recommendations
    BuyerAI-->>Buyer: Present actionable insights
    Note over Buyer,PubMCP: Process takes minutes

Getting Started

Each tool specification in this repository includes:

• Detailed API documentation
• Request and response formats
• Integration guides for different client types
• Example implementations

Contributing

We invite publishers, DSPs, and buyers interested in shaping the future of programmatic collaboration to review our specifications and contribute to building the next generation of programmatic communication.

License

[License information]