Agentic AI is moving beyond individual automated tasks towards workflows that can interpret objectives, gather information and take action. In advertising, that could mean investigating a delivery issue, activating an audience segment or answering a campaign question without requiring someone to navigate multiple dashboards.
Adoption remains early. IAB UK’s The State of AI in Advertising report found that just 4% of marketing operations are currently agentic by design, while 74% of members are experimenting with the technology.
For agents to do more than answer questions, they need access to the right tools and enough context to use them effectively. They also need clear boundaries around what they can access and change. The challenge is building connections that make action possible without losing control over how it happens.
Moving from isolated automation to connected workflows
Traditional automation follows predefined rules. An API integration, for example, can perform a specific action when it receives the right inputs. If information is missing or an unexpected error occurs, the process may stop and require human intervention.
An agentic workflow can investigate the problem, compare information across systems, and recommend a response. In programmatic advertising, an agent might examine a Deal Sync issue, identify an issue and help resolve it rather than simply report a failure.
The same principle could support buyers throughout a campaign, from interpreting a brief and recommending an activation strategy to identifying optimisation opportunities once it is live. For publishers, agents could help assess inventory quality and demand signals when assembling deal opportunities.
These applications depend on agents being able to work across systems. Without a consistent way to connect them, every new tool can mean another integration to maintain.
One connection, multiple capabilities
The Model Context Protocol (MCP) provides a standard way for AI assistants to discover tools and interact with external services. It establishes a connection between an assistant and the systems it can use, while the agent’s reasoning and orchestration determine how those tools are applied.
Onetag’s platform is made up of separate modules, each providing different capabilities. These components have their own underlying technologies, development teams, and release cycles, allowing them to evolve independently. However, connecting an AI assistant to each module separately could create additional integration work as the platform grows.
Onetag’s new MCP gateway addresses this by providing a single connection to the platform’s available tools. It discovers the tools exposed by configured modules and brings them together in one catalogue, then routes each request to the relevant module. As the platform evolves, new tools can become available through the gateway without requiring the entire integration to be rebuilt.
Resilience is an important part of this design. If one module becomes unavailable, its tools can be removed from the catalogue with a warning, while other modules remain accessible. An issue in one part of the platform therefore does not have to interrupt the entire connection.
Users authenticate with their existing Onetag accounts. The gateway applies their established permissions to determine which tools they can access, avoiding a separate set of access rules for AI interactions. Requests can also be monitored and traced across the gateway and individual modules, helping engineering teams investigate errors and establish where they occurred.
This approach allows each team to continue developing its own module while providing users with a consistent way to access the wider platform. Adding a module can be as straightforward as a configuration change, rather than a new integration project. Each module retains its own controls and audit trail, while requests remain attributable to the person who initiated them.
For users, the result is a simpler way to connect their preferred AI assistant to available tools. They can interact through a familiar interface while the gateway manages the underlying connections.
Autonomy still needs human oversight
Connecting an AI assistant to advertising tools does not mean giving it unrestricted authority. Actions can affect campaign spend and commercial relationships, so the level of autonomy must reflect the potential consequences.
People remain responsible for setting business objectives, defining success metrics and establishing the operational limits within which an agent can act. They also need to provide the knowledge and guidance that help agents interpret those objectives in context. Without this foundation, even a well-connected agent may make decisions that fail to reflect the organisation’s priorities.
Access permissions determine which tools are available, while approval requirements can keep consequential decisions under human control. IAB Tech Lab’s Agentic Advertising Management Protocols (AAMP) introduces approval gates and pricing verification for certain commercial actions. Onetag is now working with IAB Tech Lab to adopt its AAMP framework, building safeguards such as these into how our agentic workflows operate.These controls help make greater autonomy practical without removing human accountability.
A connection makes an action possible, but it does not determine whether that action is appropriate. That depends on the information available to the agent, the rules governing its behaviour and the ability to review what it has done.
Standards for a connected ecosystem
MCP is one part of a wider effort to make agentic advertising work across platforms. The AAMP framework addresses advertising-specific requirements, while MCP provides the underlying connectivity layer. They serve different purposes rather than competing to solve the same problem.
AAMP does this through a set of primitives: common building blocks that define how agents publish, discover, compare, negotiate and execute advertising transactions. Rather than introducing new formats, these primitives formalise agent communications using standards the industry already relies on. Advertising objects are described using AdCOM, giving agents a shared vocabulary that stays consistent from discovery through to reporting, while execution is handed to established standards such as OpenDirect, the Deals API and OpenRTB. MCP carries the request, while AAMP gives it advertising meaning.
Programmatic advertising developed through shared standards such as OpenRTB, which enables buyers, sellers, and platforms to communicate using a common framework. Agentic workflows face a similar requirement. Intelligence alone cannot overcome fragmented systems if agents lack a reliable way to discover tools, access context, and request actions.
MCP is still evolving, and implementations can vary between assistants and service providers. Authentication and tool discovery therefore need careful consideration as the technology develops. A modular architecture can help teams adapt while maintaining a consistent approach to access and monitoring.
Ultimately, agentic AI should reduce the operational effort involved in working across advertising systems, leaving people more time for strategic decisions and judgement. That depends on more than making tools available through a single connection. Agents need relevant context, clearly defined permissions, and actions that can be traced back to an accountable user.
Onetag’s MCP gateway is a practical step towards that model, connecting separate platform capabilities through a common interface while preserving the controls each module requires. As agentic workflows become more capable, that combination of connectivity and oversight will be essential to making them useful across the advertising ecosystem. That is why we are working with IAB Tech Lab to adopt AAMP and help shape the future of agentic advertising through open, shared standards.
Alexandra Balashoiu
Vice President of Product
Onetag
Originally Published on: IAB Tech Lab
