MCP for AI agents

Coming soon. The design is in the proposal.

AI agents are starting to draft and update content. Given raw access to a CMS, an agent can produce layouts your design system doesn't allow, skip a required field, or publish content nobody has approved. Inka's Model Context Protocol (MCP) server gives agents the same way in as people, and nothing more.

• Only what your design system allows. An agent is offered only the blocks and fields that are valid where it is working, from the same block schemas your editors use.
• Markdown in, design-system blocks out. Agents write markdown, with directives that map onto your blocks, and Inka handles how the page is stored.
• The same advice editors get. An agent has the design system's guidance on how and when to use each block, the same advice editors see while they work.
• The same advice editors get. An agent has the design system's guidance on how and when to use each block, the same advice editors see while they work.
• It sees the page, not just the data. An agent can see what each block looks like and preview the rendered page, so it can judge layout and visual impact, not only the words.
• Read, edit, preview, commit. An agent reads the page with the identifiers its changes will target, previews the result, and then commits it.
• The same checks and reviews. Everything an agent writes goes through the same rules and checks as an editor's work, and the same workflow and sign-off. An agent can speed up the work, but it can't publish content that hasn't been reviewed.
• The same record. An agent's changes are recorded like anyone else's.

The MCP server is a plugin installed in Inka, alongside the content store adapters.