MCP lets a compatible AI assistant work with information and tools from Adpulse. This guide explains what that means, how to get connected, and how to use the prompts in this library.
What is MCP?
MCP stands for Model Context Protocol. It is a standard way for an AI assistant to connect to another service and use the tools that service makes available.
Think of it as a connection between your AI assistant and Adpulse. Instead of manually copying every report into a conversation, you can ask the assistant to retrieve available account data through the connection and use it to help with a task.
The AI assistant interprets your request. The Adpulse MCP connection provides access to supported tools and data. What the assistant can actually do depends on the connection, your permissions, the advertising platform, and the tools available in that conversation.
What can I use it for?
The prompts in this library give your assistant a structured way to investigate a specific problem. For example, you can ask it to:
Investigate a possible conversion-tracking problem.
Compare performance before and after a budget increase.
Review search terms and draft negative keyword recommendations.
Investigate landing pages and conversion paths.
These are examples of tasks, not a promise that every required system or data point is accessible through Adpulse. Each prompt explains the inputs and access it needs.
What is the difference between MCP, a prompt and a skill?
MCP is the connection. It lets your assistant use the tools exposed by Adpulse.
A prompt is the instruction. It tells the assistant which task to carry out, what to check and how to present the result.
A skill is a reusable workflow. Some AI products let you save or install instructions as a skill. Installation and behaviour vary by product.
The articles in this library provide copy-and-paste prompts. You do not need to install a separate skill to use them.
Before you start
An Adpulse login with access to the accounts you want to review.
An AI assistant that supports the Adpulse MCP connection. Our setup guide covers Claude, ChatGPT and Gemini; availability and setup requirements depend on the AI product and plan.
Permission from your organisation to use the chosen AI service with your account data.
A clear task, account or campaign scope, and reporting dates.
Some investigations also need exports or manual evidence from GA4, Google Tag Manager, your CRM or call-tracking service. Adpulse access does not automatically include those records. Reviewing live pages also requires browsing capability in your AI assistant, or page content supplied by you.
Connect your AI assistant to Adpulse
In Adpulse, go to Settings > Integrations > MCP Servers.
Select Add Server and choose your AI client.
Follow the MCP connection setup guide to enter the connection details in your AI client's connection settings and complete authorisation.
Enable the Adpulse connection in the conversation you will use. Some assistants require you to enable a connection for each chat.
Keep connection credentials private. Enter them only in the intended connection settings. Do not paste passwords, API tokens or client secrets into a chat message, a prompt or a support article.
Check the connection
Start with a small, read-only request for an account you recognise. Replace the bracketed details in this example:
Using the Adpulse connection, check whether you can access [account name and account ID]. Confirm the account name, advertising platform, currency and timezone where available. Do not retrieve detailed performance or personal data, and do not change anything. If access is unavailable, explain what you could verify and what I need to check next.
Confirm that the returned account is the one you intended. If the connection does not work, check that it is enabled in this chat, that authorisation is complete, and that your Adpulse user can access the account. Do not share your credentials with the assistant to troubleshoot it.
How to use this prompt library
Choose a task. Read what the prompt does, when to use it and what information it needs.
Start a conversation with Adpulse enabled. If a task needs evidence that Adpulse cannot provide, supply an appropriate export or manual check in line with your organisation’s policies.
Copy the entire prompt. Replace the square-bracketed placeholders with your business details, account and campaign scope, and dates.
Provide missing evidence. If the assistant cannot access something, give it an appropriate export or ask for the next manual check. An unavailable record is not proof something is broken.
Review the findings. Check the sources, dates, calculations, assumptions and unanswered questions before acting.
For example, if conversions have suddenly dropped, start with the conversion-tracking investigation. Explain which conversion you expect, when the drop started, and which systems the assistant can inspect.
What should I expect back?
A useful result explains what was checked, what the evidence supports, what remains uncertain and what to do next. Depending on the task, it may include a comparison, a prioritised review or a list of verification steps with suggested owners.
The assistant may ask questions before it can reach a conclusion. That is better than a confident answer based on the wrong account, incomplete data or an assumption about your business.
Limits and safety
The initial diagnostic prompts are written to review and recommend only. They tell the assistant not to change settings, budgets, bids, keywords, negatives or automation, and not to submit leads, calls, messages, orders or test conversions.
A read-only instruction is not an access control. The Adpulse MCP connection can expose actions as well as data. Use the permissions and approval controls available in your tools, and do not approve an unexpected change request.
Data may be incomplete or delayed. Ask the assistant to state freshness, reporting coverage and conversion-lag limitations.
AI findings need review. Do not treat a suggested cause as a confirmed fault without supporting evidence.
Share only what is needed. Avoid unnecessary personal information and follow your organisation's data-handling rules.