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

# AI content creation

> Use AI to draft, edit, migrate, and operate on content via the content-management MCP server

Beyond serving content to client apps, Metabind lets AIs read and write content directly. The content-management MCP server exposes a set of tools that an AI can call to discover content types, query entries, draft new content, and update existing entries — under the same governance and audit trail as everything else.

## What's exposed

The content-management MCP server organizes its tools into three groups:

| Group         | Tools                                                                                        | Purpose                                   |
| ------------- | -------------------------------------------------------------------------------------------- | ----------------------------------------- |
| **Context**   | `set_project`, `get_current_project`, `list_projects`                                        | Tell the AI which project it's working in |
| **Discovery** | `search_content_types`, `get_content_type`, `search_content`, `get_content`, `search_assets` | Read-only inspection of project content   |
| **Creation**  | `create_content`, `update_content`                                                           | Drafting and editing entries              |

A connected AI host (Claude Desktop, ChatGPT, etc.) calls these tools the same way it calls Metabind's MCP App tools — JSON-RPC, schema-validated, audited.

## Use cases

**Drafting product descriptions with AI.**
The AI reads existing product entries, drafts copy in your house voice, creates a draft entry. A human editor reviews and publishes.

**Bulk migration from another CMS.**
A planning agent maps fields from the source CMS to your Metabind content type schema. A worker agent iterates source records and creates Metabind entries via `create_content`. Errors surface in the audit log for review.

**Content operations.**
"Tag every article published before 2024 as `archive-candidate`" or "Find any entries missing a hero image and flag them." Reads + filtered updates on a schedule, with audit trail for compliance.

**Editor assistance.**
An editor in a custom in-app assistant asks the AI to "summarize this article in 3 sentences" or "rewrite the intro to match the tone of last month's post." The AI uses content tools to read context and update the draft.

## Connecting

The content-management MCP server is exposed at:

```
https://mcp.metabind.ai/cms
```

Configure it in your MCP host the same way you configure project-scoped MCP Apps:

```json theme={null}
{
  "mcpServers": {
    "metabind-cms": {
      "url": "https://mcp.metabind.ai/cms",
      "headers": {
        "Authorization": "Bearer mb_..."
      }
    }
  }
}
```

The token's scope determines what the AI can access — at the moment, it inherits the user's organization role and the projects that role gives access to.

## Governance and safety

All the platform-wide governance applies:

* **Schema validation.** Every `create_content` and `update_content` call is validated against the content type's schema before writing.
* **Permissions.** The AI acts under the user's role. A Viewer-scoped token can read but not write. An Editor-scoped token can draft. Publishing remains a human action by default.
* **Audit trail.** Every call appears in the audit log with input, output, and the user identity behind the AI host.

For projects in regulated industries: consider keeping AI tokens scoped to **Editor (drafts only)** and requiring a human to publish.

## Integrating into a custom assistant

If you're building a custom assistant via the [Assistant SDK](/guides/getting-started/embed-an-assistant), connect both your project's MCP App server *and* the content-management MCP server. The AI will have tools for both rendering UI (Interactive Tools) and reading/writing content (CMS tools) in one conversation.

```ts theme={null}
const assistant = createAssistant({
  project: { url: "https://mcp.metabind.ai/my-org/my-project", token: ... },
  additionalServers: [
    { url: "https://mcp.metabind.ai/cms", token: ... }
  ],
  llm: anthropic({ ... })
});
```

## Related

<CardGroup cols={2}>
  <Card title="Managing content" icon="folder-open" href="/content/managing/overview">
    Editor workflow in MCP App Studio.
  </Card>

  <Card title="Querying content" icon="magnifying-glass" href="/content/querying/overview">
    Read-only access via REST, GraphQL, SDKs.
  </Card>

  <Card title="Audit logs" icon="clipboard-list" href="/guides/operations/audit-logs">
    Where AI content actions are recorded.
  </Card>

  <Card title="Tools and Types" icon="wrench" href="/guides/concepts/types">
    The MCP App platform's tool model.
  </Card>
</CardGroup>