A marketing audit API lets your software request a website review and retrieve the findings for your team or AI agent to inspect. The useful workflow is not “audit, then change everything.” It is request, wait, check the evidence, and propose work someone can review.
Magister announced its REST API on November 12, 2026. It includes a marketing-audit endpoint that accepts a public-site URL and returns results through a background job. You can request that standalone analysis without creating a brand first, though API authentication is still required.
For founders and marketers, the question is where that analysis belongs: inside an assistant you already use, or inside software your team builds.
Choose the API or your existing assistant
Use a REST API when you want your own application or script to submit the request and handle the result. You are building the connection between the audit and whatever happens next.
If you want to work inside Claude, ChatGPT or Codex instead, Magister also documents an MCP route. The Model Context Protocol architecture describes how AI applications connect to servers that expose tools and context. With Magister’s Connect route, your assistant supplies the conversation and model; Magister supplies the brand workspace and marketing tools.
These are different ways to work, not interchangeable credentials or billing arrangements. Choose the route that fits the task. You do not need a custom integration just to ask an assistant for marketing help.
Start with one public website
Magister’s marketing-audit request is POST /v1/audits/site. Its required JSON field is url, containing the public-site URL you want reviewed.
The documented result includes an overall score, scores by area, findings and opportunities. That is the analysis you will inspect—not a promise that the endpoint will publish fixes or execute a marketing plan.
A domain-only request also does not establish access to private analytics, advertising or CRM accounts. Keep public-site observations separate from conclusions that require authorized account data.
For an initial integration, we recommend keeping the scope narrow: one site, one completed audit, and a proposed task list. Establish that the handoff works before adding recurring requests or more sites.
Keep the API key on your server
The API uses bearer-key authentication. Starting a marketing audit requires a full API key; a read-only key cannot start paid work.
Keep that key in a server-side secret store or environment variable, as Magister’s authentication documentation directs. Do not put it in a public repository or browser-delivered page. The person reviewing a report does not need the secret that started the job.
Before submitting, decide where your integration will store the job ID and how it will handle failure. Keep the request and its result connected so your agent does not mistake an older report for a newly completed review.
Wait for completion, not just acceptance
The audit endpoint returns HTTP 202 and a job_id. That means the request was accepted, not that the audit is finished. MDN’s explanation of 202 Accepted makes the distinction explicit: processing may not have started, and acceptance does not guarantee success.
Poll GET /v1/jobs/{job_id} using the key that started the job. Wait for the returned poll_after_seconds between checks rather than repeatedly asking as fast as possible. Polling is free.
A job moves from running to succeeded or failed. Only hand a succeeded result to your agent as a completed audit. If it fails, inspect the error instead of asking the agent to fill in missing findings.
For an event-driven integration, you can provide a public HTTPS webhook_url to receive success or failure notifications. Verify the signature using the documented signing procedure. Deliveries can repeat, so make your handler idempotent: receiving the same event again should not create another set of tasks.
Turn findings into proposed work
Do not stop at the overall score. Read the findings and check what evidence supports each one. Where a conclusion needs information the audit does not provide, record that gap rather than treating the conclusion as settled.
Our recommended handoff to an agent includes:
- The website and date of the review.
- The relevant finding and its supporting evidence.
- Your intended audience, offer and current constraints.
- The information still needed before making a change.
- A request for proposed work, not immediate publication.
These are suggested handoff fields, not a claim about the audit’s response schema or an automatic task-creation feature.
Ask your agent to separate an observed issue from a hypothesis about its business effect. A proposed copy change is not evidence that conversions will rise. A missing detail on a page is not proof that it caused lost sales.
For each task you accept, specify what will change and how you will verify it. Keep implementation and publication as separate decisions from obtaining the analysis.
Understand API credits before adding volume
Magister’s public API uses prepaid credits billed separately from Connect and Agent plan credits. Do not assume a subscription’s monthly allowance pays for REST API requests.
The billing model distinguishes a pre-start credit hold from the actual usage charged for a run. Available balance must accommodate the hold, not just an expected charge. Record the completed job’s credits_charged rather than treating an estimate as the final bill.
The optional max_credits field is an admission check: it rejects a job before starting if the required hold exceeds that value. Do not describe it as a guaranteed ceiling on the final charge.
A request refused before it starts costs nothing. A job that fails partway can still incur a charge for the work performed. Check those conditions before building retries into your integration.
Review the changes the audit prompts
An audit can inform a content brief, but it does not remove the need to check the resulting copy. Verify product details, offers, links and metadata before publishing anything your agent drafts.
Google’s guidance on generative AI content emphasizes accuracy, quality and relevance, including review before publication. Apply that discipline to the work an automated audit prompts, not only to the audit report itself.
Keep the original finding alongside the proposed edit. That gives the reviewer something concrete to evaluate: does this change address the observed issue, and are its claims supported?
Build the smallest complete workflow first
Start with a finished result, not a successful submission: request the audit, wait for its outcome, inspect the findings, and give your agent enough context to propose useful work.
Then review that work before making live changes. The endpoint supplies analysis; your integration needs a deliberate handoff from evidence to action.
Read the marketing-audit endpoint reference for the request and response details, then use the API quickstart to set up authentication and understand the shared job workflow. Its worked example covers AI visibility, so use the audit reference for this post’s endpoint and fields.