If you have started seeing terms like MCP, tool calling, or OAuth scopes appear in marketing conversations, you are not alone. This MCP advertising glossary and AI ads terminology reference exists precisely because these concepts are crossing from developer documentation into everyday marketing practice, fast. Tool calling in AI advertising means an AI assistant executing a specific action against a real platform, such as pulling campaign performance data or adjusting a budget, rather than simply generating text. This glossary defines 30 terms in plain marketing language, so you can follow the conversation and make informed decisions about AI-powered advertising tools.
Why Marketers Need an MCP Vocabulary Now
A year ago, most marketers had no reason to understand how AI assistants connect to external tools. That changed when platforms like Claude, ChatGPT, and Gemini started gaining real integrations with advertising accounts. The infrastructure making those integrations possible, primarily the Model Context Protocol, introduced a layer of technical vocabulary that is now showing up in product documentation, agency pitches, and vendor comparisons.
Understanding these terms is not about becoming a developer. It is about knowing what a vendor actually means when they say their tool supports “full write access via MCP” or “scoped OAuth authentication”. The difference between those claims can determine whether your AI assistant can pause a campaign or only read its data.
Fluency in AI agent marketing vocabulary is becoming the same kind of professional baseline that understanding UTM parameters or conversion tracking once was.
The Core MCP Advertising Glossary: 30 Terms Defined
1. MCP (Model Context Protocol)
An open protocol developed by Anthropic that defines how AI assistants communicate with external tools and data sources. MCP standardizes the way an AI model sends requests to a server and receives structured responses. In advertising, MCP is the plumbing that allows an AI assistant to actually interact with your Google Ads or Meta Ads account rather than just talking about it.
2. MCP Server
A backend service that implements the Model Context Protocol and exposes a set of tools to an AI assistant. An MCP server sits between the AI model and the advertising platform. When you connect Claude to an advertising tool, it communicates through that tool’s MCP server. The server handles authentication, translates requests into platform API calls, and returns results.
3. MCP Client
The AI assistant or application that initiates requests to an MCP server. Claude.ai is a common MCP client. When a user types a request in Claude, the client identifies which MCP server and which tool to invoke, sends the request, and presents the result back to the user.
4. Tool Calling
Tool calling is the mechanism by which an AI model invokes a specific function exposed by an MCP server. In advertising terms, a tool call might look like: get_campaign_performance(campaign_id=123, date_range=”last_7_days”). The AI does not browse a dashboard. It sends a structured call to a defined tool, receives structured data, and interprets it. This is fundamentally different from web browsing or reading a screenshot.
5. Tool Definition
A structured description of what a tool does, what parameters it accepts, and what it returns. The AI model reads tool definitions to know which tool to use for a given request. A well-written tool definition for advertising might specify that a tool accepts a campaign ID and a date range, and returns impression, click, and cost data. Poor tool definitions lead to incorrect tool selection by the AI.
6. Tool Confirmation
A user-facing approval step that appears before a write action is executed. When an AI assistant is about to create a campaign, modify a budget, or pause an ad set, tool confirmation shows the user exactly what will happen before it happens. This is a safety mechanism, not a limitation. Some MCP implementations require confirmation for every write action; others skip it entirely, which introduces risk in advertising contexts where unreviewed changes can have immediate budget consequences.
7. Read-Only Tool
A tool that retrieves data without modifying anything. Pulling a keyword performance report is a read-only action. Many early MCP implementations for advertising are read-only, meaning the AI can analyze but cannot act. This distinction matters when evaluating what a given MCP server actually enables.
8. Write Action
Any tool call that modifies data on a connected platform. Creating an ad group, adjusting a bid, updating a headline, or pausing a campaign are all write actions. In advertising, write actions have direct financial and performance consequences, which is why confirmation steps and access control matter considerably more than they do in read-only integrations.
9. Envelope
In MCP architecture, the envelope refers to the structured wrapper around a tool call or response. It contains metadata such as the tool name, parameters, request ID, and protocol version alongside the actual payload. Marketers rarely interact with envelopes directly, but understanding that every tool call is a structured message with a defined format helps explain why MCP integrations behave consistently across different tools and platforms.
10. OAuth
An authorization framework that allows a user to grant a third-party application access to their account on another platform without sharing their password. In advertising, OAuth is how you securely connect your Google Ads or Meta Ads account to a tool. The user approves the connection in the platform’s own interface, and an access token is issued to the requesting application.
11. OAuth Scopes
OAuth scopes define the specific level of access granted during an OAuth connection. A scope might grant read access to campaign data, or write access to ad creative, or billing access to budgets. Scopes are granular by design. When evaluating an MCP advertising tool, the scopes it requests during setup tell you exactly what it can and cannot do on your behalf. An integration that requests only read scopes cannot create or modify campaigns, regardless of what its marketing page claims.
12. Access Token
A credential issued after a successful OAuth authorization. It represents the permission the user granted and is used by the MCP server to authenticate API requests to the advertising platform. Access tokens are typically time-limited and may be refreshed automatically using a refresh token.
13. API Key
A unique identifier used to authenticate a user or application against a service’s API. In MCP advertising tools, an API key is often used to authenticate the connection between the AI client and the MCP server itself, separate from the OAuth tokens used to access individual ad platforms. For example, you might use an Adsroid API key to authenticate Claude’s connection to the Adsroid MCP server, while separate OAuth tokens handle the Google Ads and Meta Ads connections.
14. Endpoint
The URL address of an MCP server that an AI client connects to. For example, a single endpoint URL can expose all tools available on that server. The endpoint is what you enter when configuring a custom MCP connector in an AI assistant. One endpoint can surface dozens or hundreds of individual tools depending on how the server is built.
15. org_id (Organization ID)
A unique identifier for an organization within a platform or MCP server. In multi-account advertising environments, the org_id is what allows a single authentication context to access multiple client projects without mixing data between them. When an AI assistant resolves which account to act on, the org_id is part of the routing logic. This is particularly relevant for agencies where one workspace may contain dozens of client accounts.
16. Project (in MCP Context)
A discrete unit within an MCP server that contains a specific set of connected accounts, credentials, and business context. Projects are how multi-client isolation works. When each client lives in its own project, their ad data, business identity, and account connections remain separate even when managed through the same API key or conversation session.
17. Business Context
A structured data layer that carries information about a business, such as its offer, target audience, positioning, unique selling points, and customer pain points, into the AI assistant’s working context. In advertising, business context is what separates AI-generated ad copy that reflects your actual brand from generic output that could apply to any competitor. Some MCP servers load business context automatically before any tool is called; most do not include this layer at all.
18. Context Window
The maximum amount of text and data an AI model can process in a single interaction. In MCP advertising workflows, the context window determines how much account data, tool results, and conversation history the AI can hold simultaneously. Large context windows allow more comprehensive analysis across multiple campaigns or accounts in one session.
19. Prompt
The input a user provides to an AI assistant to initiate a task. In advertising workflows, a prompt might be: “Analyze my top five Google Ads campaigns by ROAS over the last 30 days and recommend budget reallocation.” The quality of the prompt, combined with the tools available to the AI, determines the quality of the output.
20. System Prompt
A set of instructions provided to the AI model before the user interaction begins. MCP servers can inject system-level context, such as business information or behavioral rules, through mechanisms that function like system prompts. This is how some implementations ensure the AI behaves consistently across sessions without the user needing to re-explain their business every time.
21. Agent (AI Agent)
An AI system that can take sequences of actions autonomously using available tools. An AI agent for advertising does not just answer questions. It can retrieve data, analyze it, decide on an action, execute that action, check the result, and continue. The degree of autonomy versus human confirmation varies by implementation and is a critical governance consideration for marketing teams.
An AI agent that can read your campaign data but cannot modify it is a reporting assistant. An AI agent with write access is an operator. The distinction has real consequences for budget control and account safety.
22. Agentic Workflow
A multi-step task sequence executed by an AI agent using tool calls. In advertising, an agentic workflow might involve pulling keyword data, identifying underperforming terms, generating negative keyword recommendations, adding those negatives to the correct ad group, and confirming the change, all within a single conversation. Understanding how AI-native paid acquisition workflows operate helps teams design processes that use these capabilities safely.
23. Tool Chaining
The practice of using the output of one tool call as the input for another. For example: call a tool to get campaign IDs, then call another tool to get performance data for each ID, then call a third tool to update bids based on that data. Tool chaining is what enables complex multi-step advertising tasks within a single AI session.
24. Paused State
A default condition applied to newly created advertising elements through an MCP write action. Rather than launching a new campaign, ad set, or ad immediately upon creation, responsible MCP implementations create them in a paused state so the user can review before anything goes live. This is a standard safety practice in AI-driven campaign creation.
25. Zero Data Retention
An architecture model in which no advertising account data is stored on the MCP server’s infrastructure after a tool call resolves. Every request pulls live data from the connected platform in real time and returns it directly to the AI session. Zero data retention reduces the risk associated with data breaches and simplifies compliance, since no historical ad data accumulates on a third-party server.
26. Multi-Account Access
The ability for a single authentication context, such as one API key or one login, to access multiple advertising accounts or client projects. For agencies, multi-account access is essential. The key question is whether accounts are properly isolated from each other or whether data from one client could appear in another client’s session.
27. GEO (Generative Engine Optimization)
The practice of optimizing content so that it is cited and referenced by AI search engines such as ChatGPT, Claude, Gemini, and Perplexity when answering user queries. GEO is becoming a distinct discipline alongside traditional SEO. Understanding how to identify which sources AI search engines cite is increasingly relevant for marketers building content strategies in an AI-first search environment.
28. Ad Creative Generation
The use of AI to produce advertising creative assets, including headlines, descriptions, images, and copy variations, based on business context and campaign parameters. When this happens through an MCP tool call rather than a standalone AI chat, the output can be directly applied to a campaign rather than copy-pasted manually. The quality difference between generic AI creative and context-grounded creative is significant.
29. Competitor Intelligence Tool
An MCP tool that retrieves data about competitor advertising activity, such as which ads a competitor is running, on which platforms, and with what messaging. In the MCP advertising vocabulary, this is distinct from ad performance tools that read your own account data. Competitor intelligence tools pull external data and are typically read-only by nature.
30. Reach Estimation
A tool call that returns a projected audience size for a given set of targeting parameters before a campaign is launched. Meta Ads surfaces this data through its API, and MCP servers with deep Meta integration can expose reach estimation as a discrete tool. This allows an AI assistant to validate audience size during campaign planning rather than requiring the user to check manually in Ads Manager.
How These Terms Come Together in Practice
Reading these definitions in isolation is useful, but seeing how they interact clarifies why the vocabulary matters. Consider a typical AI-assisted campaign workflow:
- A marketer connects their Meta Ads account to an MCP server via OAuth, granting specific scopes for campaign reading and writing.
- They open Claude and send a prompt asking for an analysis of last month’s ad set performance.
- Claude identifies the right tool, sends a structured tool call to the MCP server’s endpoint, and receives live data.
- The AI analyzes results and recommends creating a new ad set targeting a lookalike audience.
- Before executing, Claude surfaces a tool confirmation showing the exact parameters of the new ad set.
- The marketer approves. The ad set is created in a paused state.
- The marketer reviews it in Ads Manager and activates it manually.
Each step in that workflow involves specific terms from this glossary. Understanding them means understanding what the tool is actually doing at each stage, and where the risks and control points are.
Where Adsroid MCP Fits
Adsroid MCP is a Model Context Protocol server that is part of the broader Adsroid platform. It connects AI assistants like Claude directly to Google Ads, Meta Ads, Google Analytics 4, Google Search Console, and Ad Radar through a single endpoint. It covers over 140 tools across four action types: Analyze, Manage, Create, and Test.
Several terms in this glossary describe features that distinguish different MCP implementations. Adsroid MCP includes business context loading, which means the AI automatically receives the user’s business identity before acting on any tool. It uses tool confirmation for every write action. New campaigns and ads are created in a paused state. It operates on a zero data retention model. And it supports multi-account access with full project isolation, which is relevant for agencies managing multiple clients.
For teams evaluating MCP advertising tools, the Adsroid MCP documentation is a useful reference for understanding how these architectural terms translate into concrete product behavior. The difference between an MCP server that claims write access and one that implements it with proper confirmation, isolation, and safety defaults is significant, and this glossary gives you the vocabulary to ask the right questions.
Governance is also worth understanding in depth. Enterprise marketing teams using MCP for ad account governance need to think carefully about confirmation steps, org isolation, and permission models, all of which map directly to terms covered here.
Frequently Asked Questions
What does tool calling mean in AI advertising?
Tool calling means an AI assistant invokes a specific function against a connected platform rather than generating text responses only. In advertising, a tool call might retrieve campaign data, update a bid, or create an ad set. The AI sends a structured request to an MCP server, which executes the action via the platform’s API and returns a result.
What is the Model Context Protocol (MCP)?
MCP is an open protocol developed by Anthropic that standardizes how AI assistants communicate with external tools and data sources. It defines the message format, tool description structure, and interaction model between an AI client and an MCP server. In advertising, it is the infrastructure that enables AI assistants to take real actions in platforms like Google Ads and Meta Ads.
What is the difference between read-only and write access in MCP?
Read-only access allows an AI assistant to retrieve data without modifying anything. Write access allows it to create, update, or delete elements in the ad platform. Many MCP advertising servers are read-only. Write access is what enables campaign creation, budget changes, and bid adjustments, but it also requires proper confirmation steps and governance to prevent unintended changes.
What are OAuth scopes in the context of advertising MCP tools?
OAuth scopes define the specific permissions granted when you connect an advertising account to an MCP tool. A scope might cover reading campaign performance, modifying bids, or accessing billing data. The scopes an MCP server requests during setup are a reliable indicator of what it can actually do on your account. Narrow scopes mean limited access; broad scopes mean the tool can perform a wider range of actions.
What is business context in AI advertising tools?
Business context is a structured data layer that gives the AI assistant information about a specific business, including its offer, target audience, positioning, and unique selling points, before it generates ad copy or recommendations. Tools that include business context produce output grounded in the actual brand. Tools without it produce generic results that require significant manual editing.
What is zero data retention in MCP advertising?
Zero data retention means the MCP server does not store advertising account data after a tool call completes. Every request resolves in real time against the connected platform, and no historical data accumulates on the MCP server’s infrastructure. This reduces third-party data exposure and simplifies compliance for teams handling sensitive account information.
What is org_id and why does it matter for agencies?
org_id is an organization identifier used within MCP server architecture to route requests to the correct account or project. For agencies managing multiple clients, org_id is what enables multi-client access from a single authentication context while keeping each client’s data isolated. Without proper project isolation, data from one client could potentially surface in another client’s session.
What is the difference between an MCP server and an API?
An API is a general interface for accessing platform functionality programmatically. An MCP server is a specific implementation that exposes functionality through the Model Context Protocol so that AI assistants can discover and call tools in a standardized way. APIs require developers to write code to interact with them. MCP servers make that functionality directly available to AI models without additional coding.
What is a paused state in AI campaign creation?
When an AI assistant creates a campaign, ad set, or ad through an MCP write action, a paused state means the new element is created but not immediately active. The user can review the campaign in the ad platform before enabling it. This is a standard safety default in responsible MCP implementations and prevents untested campaigns from spending budget without human review.
