Getting Started

This page takes you from nothing to an agent that answers "what should we write next?" with your own data.

The short way: let your agent set it up

After you register, paste this line into Claude, ChatGPT or any MCP client:

Read https://kafkai.com/docs/onboarding.md and set up Kafkai for me.

The agent reads the agent onboarding page, gives you the connection steps for your client, and then creates your first project with you. It tells you the credit cost before every paid call.

If your agent cannot open web pages, follow the steps below instead.

1. Create a Kafkai account

Register here and verify your email address. Tool calls that return data need credits, so check your balance on the dashboard and buy a pack from the pricing page if it is empty. We sometimes grant free credits to new accounts; when we do, we tell you how many and how long they last.

2. Find your dashboard

After signing in, the dashboard shows three things you need:

the MCP server URL to paste into your client,
your API tokens (for clients that use header authentication),
your credit balance and recent usage.

3. Connect your agent

Pick the guide for your client:

All of them end the same way: your agent lists Kafkai tools whose names start with kafkai_.

4. Create your first project

You can create a project directly from a conversation with your agent:

Create a Kafkai project for https://example.com targeting Japan, and add https://competitor-a.com and https://competitor-b.com as competitors.

Name the country. A project is locked to one country's search results, so Kafkai will not guess one — leave it out and the call fails for free, listing the markets you can pick.

The agent calls kafkai_create_project, and Kafkai starts the keyword research in the background. Research usually takes a few minutes. Ask the agent to check progress:

Show me a summary of the example.com project.

When the keyword counts stop at zero, the research is still running — ask again a little later.

5. Ask what to write

Once research is done, useful first prompts look like this:

Which keywords should we write about next for example.com, and why?

Pick the best Catch-up keyword for example.com, research it on the web, and draft an article for it.

How does competitor-a.com beat us, and what should we publish to close the gap?

Behind these answers, the agent chains Kafkai tools on its own: list projects, read the project summary, pull keywords for the recommended strategy, then search the web and write. Each tool response includes a next_step_suggestion field that guides the agent to the next call, so you do not need special prompting.

More prompts, each with the tools the agent calls and what comes back, are on the Example Prompts page.

6. Ask what changed

Kafkai dates every rank it measures, so your agent can compare a keyword's position today against where it sat months ago.

Rankings refresh only when you ask for them, so run an update when the data is stale:

Update the keyword rankings for example.com.

A project needs at least two updates before there is anything to compare. After that, these prompts work:

What moved for example.com in the last three months? Which keywords did we lose?

Where is competitor-a.com losing positions, and which of those keywords should we write about?

Both go to kafkai_get_ranking_changes, which sorts the biggest moves first and reports the date of every position it compares.

Costs

Every data-returning tool call spends credits from your prepaid balance. Checking the balance itself is free — ask the agent "how many Kafkai credits do I have left?" at any time. Details are on the Credits and Billing page.