Help & Documentation

Understanding MCP and DashHarbor

Learn how native EmDash MCP connects AI tools to your website, what they can do and how you control their access.

MCP gives an AI application a standard way to work with another service. On DashHarbor, it lets a compatible AI assistant or development tool connect directly to your website's native EmDash content system.

You can use it to inspect your content, understand its structure or help prepare changes. You choose the tool, the website and the permissions it receives.

What MCP does

The Model Context Protocol, usually shortened to MCP, is an open protocol for connecting AI applications to tools and information. It gives applications a shared connection method instead of requiring a different custom integration for every service.

Your AI application is the client. Your website's EmDash installation provides the server, which makes a set of tools available. The client can ask those tools to read content or carry out permitted actions, then use the results in your conversation or workflow.

MCP is the connection, not the AI model. You still use your chosen AI application's own account and service.

What an AI tool can do with your website

The available EmDash tools and your permissions determine what is possible. Common tasks include:

  • Reading existing pages and other content.
  • Inspecting collection schemas to understand how content is organized.
  • Finding content that needs attention.
  • Creating and editing content when you deliberately allow write access.
  • Reading, uploading or updating media when the corresponding permissions are granted.

Write access needs care. An owner's `content:write` permission can allow publication and deletion as well as creation and editing. It is not draft-only access. Asking an assistant to prepare drafts does not reduce the authority of its token.

Start with read access when you want an assistant to review your website. Add permissions only for tasks you intend it to perform, and review proposed changes in EmDash.

How DashHarbor provides MCP

DashHarbor uses the MCP functionality built into EmDash. There is no separate Codex-specific CMS or content store. A compatible client connects to your website's permanent DashHarbor address, ending in /_emdash/api/mcp.

Each website has its own endpoint and API tokens. A connection to one website does not grant access to another. Separate production and staging installations also have separate identities and credentials; the DashHarbor application staging environment is not a copy of your website.

Create and manage your personal access tokens in the website's native EmDash editor. Choose the token's permissions and expiry, and revoke it when access is no longer needed. Your native EmDash role and content permissions still apply.

Can I use my preferred AI application?

You are not tied to one AI provider, but an application must support this particular connection: remote MCP over Streamable HTTP with native bearer-token authentication. General MCP support alone does not establish compatibility.

Codex CLI is the existing tested example. Tools customers may also consider include ChatGPT, Claude, Claude Code, Cursor, Visual Studio Code, GitHub Copilot, Windsurf, Cline, Zed, Continue and JetBrains AI Assistant. These other clients have not been verified against this DashHarbor connection. In particular, we do not promise ChatGPT or Claude compatibility based on general MCP support.

Compatibility also depends on your website's installed EmDash release and managed configuration. An endpoint being visible is not proof that a client connection has been tested. Check the available compatibility information before setting up access.

Choose your next step

Connect AI tools to your DashHarbor website for the common setup steps, permissions and the Codex connection example.

If you already have a connection, disconnect AI tools safely to revoke access and remove local credentials.