> ## Documentation Index
> Fetch the complete documentation index at: https://docs.seosorted.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP overview

> Connect seosorted.ai to Claude and other AI agents, so they can research keywords, write articles and publish for you.

The SeoSorted MCP server lets an AI agent work inside your workspace. Instead of you opening
the dashboard, the agent calls SeoSorted directly — looking up keyword data, reading your
latest audit, generating a draft, checking how it performed.

It speaks [Model Context Protocol](https://modelcontextprotocol.io), so it works with Claude
and any other MCP-capable client.

<Note>
  Connecting takes one click: paste the server URL into your client, sign in the way you always
  do, and approve. There is no API key to create or copy — see [Connecting](/mcp/connecting).
</Note>

## What an agent can do

<CardGroup cols={2}>
  <Card title="Research" icon="magnifying-glass">
    Look up live volume, difficulty and CPC for new terms, and browse the keywords you already track.
  </Card>

  <Card title="Audit" icon="stethoscope">
    Run a site audit and read the findings grouped by severity, then drill into specific issues.
  </Card>

  <Card title="Write" icon="pen-nib">
    Turn a keyword into a full draft through the same pipeline the dashboard uses.
  </Card>

  <Card title="Publish" icon="paper-plane">
    Send a finished article to your connected CMS — with your explicit approval, every time.
  </Card>
</CardGroup>

## What it looks like

A real exchange, once connected:

> **You:** Which of my keywords don't have an article yet? Write the best one.
>
> **Agent** calls `list_keywords`, sees five untouched terms, picks the one with the
> highest volume and lowest difficulty, calls `generate_article`, then `wait_for_job`
> until the draft is ready.
>
> **Agent:** Done — *"Best SEO Tools and Resources Guide"*, 1,988 words, saved as a draft.
> That used 1 of your 3 monthly article credits. Want me to publish it?

## The rules it plays by

The agent is not a way around your plan, your permissions or your credits.

<AccordionGroup>
  <Accordion title="It spends the same credits you do">
    Generating an article costs one article credit, exactly as it does in the dashboard. When
    you run out, the agent is told so and given your upgrade link — it cannot generate anyway.
  </Accordion>

  <Accordion title="It cannot exceed your plan">
    Publishing, scheduling and integrations are Pro features. On Free, those tools are visible
    but return an upgrade link instead of running.
  </Accordion>

  <Accordion title="It cannot exceed your role">
    The agent acts as you, and your workspace role is re-checked on every call. If you are
    demoted or removed, its access changes with you immediately.
  </Accordion>

  <Accordion title="It only gets what you ticked">
    You choose the workspace and the permissions on the approval screen. Publishing is off by
    default, and anything unticked is refused when used — not merely hidden.
  </Accordion>

  <Accordion title="It always asks before touching your website">
    `publish_article` refuses to run unless the agent passes an explicit confirmation, so a
    well-behaved agent shows you the article and destination first. It also has a dry-run mode
    that shows exactly what would be sent without sending it.
  </Accordion>
</AccordionGroup>

See [Plans, credits and limits](/mcp/limits) for the specifics.

## Next steps

<CardGroup cols={2}>
  <Card title="Connect a client" icon="plug" href="/mcp/connecting">
    Create a key and point Claude at your workspace.
  </Card>

  <Card title="Tool reference" icon="wrench" href="/mcp/tools">
    All twelve tools, what they cost and what they return.
  </Card>
</CardGroup>