Your agent can reason about marketing. It still needs account tools.
You can build an agent that explains campaign performance in a prototype. Making that agent useful inside a marketing team introduces a different set of tasks: account authorization, data access, execution status, repeated operations, and a record of changes.
Those tasks can consume the work you intended to spend on the agent itself. A prompt cannot supply an account permission boundary or tell you whether an asynchronous launch completed. Your integration needs an operating layer around the model.
Connect your agent to the ad account
Connect your OAuth-compatible MCP client to Superlaunch and build the agent’s workflow around the available tools. Superlaunch supplies account access checks, synced Insights, launch operations, rules, context, and Activity. You keep ownership of the model, orchestration, review policy, and evaluation.
Keep the marketer’s conversation simple while your integration handles account selection and tool calls. Try the AI agent prompts.
Build an internal acquisition assistant
Your team wants an assistant that reviews acquisition performance, proposes a creative launch, and prepares rules for a marketer to review. You run the agent in your own environment and connect it to the team’s authorized accounts.
The following is an illustrative workflow you can adapt to your account.
How this works in practice
Connect through MCP and OAuth
Use an MCP client that supports Superlaunch’s OAuth flow. Connect to https://mcp.superlaunch.pro/mcp and authorize the connection with the access the task needs. Test it by reading one account.
Your client should let the marketer choose an account by name and resolve its identifier for tool calls. Ask for clarification when names are ambiguous. OAuth scopes and the user’s account permissions both apply; handle denied access without switching to another account.
Build reports from the available account data
Discover the current tools and Insights capabilities. Read the relevant account context, choose the reporting dates and conversion event, and check the returned data coverage. Handle pagination and missing results in your integration.
A supported metric does not mean the selected account records it. Show the marketer the period and attribution settings used for each comparison, and identify gaps before making recommendations.
Preview the work and check whether it finished
Use Launcher tools to find the named creative collection and template. Check the account’s campaign and budget limits, then show a launch preview for review. After the marketer approves it, submit the launch and check its status.
A submitted launch may still be running or may fail. Retain the request identity and returned launch reference, handle uncertain responses without duplicate submissions, and report the actual result. These are integration responsibilities, not details the marketer should have to type into a prompt.
Decide which system schedules each task
Use native Launcher schedules for supported recurring creative tests and autorules for recurring account checks and actions. They can run after the agent conversation ends.
Your agent scheduler must start a weekly report job, query Superlaunch, compose the report and deliver it through your chosen channel. Use Activity and run results to investigate failures or denied actions.
AI agent prompt examples
Connect your AI agent to Superlaunch, then try these prompts in your conversation. Replace the names and example values with your own. Use account, collection and template names; the agent should find them and ask you to choose if more than one matches.
Check what your agent can do
Use Superlaunch to find [account name]. Check the available tools, saved account context and performance data. Tell me which parts of a weekly performance review and creative launch you can handle, and what setup or access is missing. Make no changes.
The integration resolves names to tool identifiers and reads current tool schemas. Marketers should not have to supply API payloads or account IDs.
Prepare a useful first task
Review [account name] for last week against the previous week. Explain the biggest changes in spend and purchases, with supporting data. Then preview a paused launch using [collection name] and [template name]. Show the budget and anything I need to review before creation.
Your client should retain account selection and tool results across the conversation. Check data coverage and attribution when producing the comparison.
Complete the launch and handle failures
Create the launch we just reviewed. Check its status and tell me what completed or failed. If you lose the response, check whether that launch already exists before trying again.
Implement request identity and status checks in your integration. A prompt alone does not prevent duplicate submissions or recover a lost response.
Plan a recurring report
Help me set up a Monday report for [account name] in our agent scheduler. Each report should compare the previous two full weeks, summarize spend and purchases, and suggest next steps. Explain any setup I need to complete. Keep the report job read-only.
Your agent scheduler must support starting the job. Superlaunch supplies the account reads; native rules and Launcher schedules handle supported recurring account actions.
Before you rely on the result
Superlaunch provides the account tools and execution controls. You host and operate your agent, manage its review flow, and handle retries and errors. Both OAuth scopes and current account permissions matter; write scope alone does not grant access to every account.
A useful first step
A useful first integration reads one account, produces a recommendation with evidence, and prepares a launch preview. Add submission only after you can handle permissions, missing configuration, pending work, and failures.
Your developers can focus on agent behavior while reusing Superlaunch’s account tools, recurring execution, permissions, and activity records.
Read the setup documentation