Your content stack, in the editor you already have open.

You live in Claude Code, Cursor or Windsurf. Content Agents speaks MCP, so the research, the writing, the publishing and the analytics are tools your agent can call — not another tab, another login and another dashboard to remember.

What changes

• The platform is an interface, not a destination — Connect once with a brand API key. From then on your agent plans, writes, runs the SEO pass, publishes and reads the analytics back — in the same session where you are shipping the product.
• More than 70 tools, not three — Not a thin wrapper over one write endpoint. Keyword research, competitor pulls, brand context, structures, enrichment, scheduling, social, newsletters and the visibility measurement are all reachable.
• Your brand rules travel with the call — Voice, terminology, banned claims and saved structures load from the brand, not from your prompt. The agent cannot quietly write off-brand because somebody forgot a line in the system message.

How it works

• Make a key — Create a brand API key in settings. It scopes to one brand, and you can hold several — one per project you run.
• Add the server — One entry in your client config. There are written walkthroughs for Claude Code, Claude Desktop, Cursor, Windsurf, VS Code with Cline and ChatGPT.
• Ask for the work — "Find the three queries we are closest to ranking for, write the one with the best shot, enrich it, and queue it for Thursday." The agent does each step as a tool call you can read.
• Keep the gate — Publishing stays an explicit action. Set the brand to propose-only and nothing goes live until a human says so — including when the caller is your own agent.

What you use

• MCP Server — The whole workspace, exposed as tools any MCP client can call.
• Editor-in-Chief — The decision layer: what to write next, and the evidence for it.
• Brand Data — Your own research and documents, grounding every draft the agent writes.
• AI Visibility — Where the answer engines cite you, per market and engine — read it back as data.
• Content Planning — A plan the agent can propose against, instead of a calendar you fill by hand.
• Brand Autonomy — The gate between proposing and publishing, set per brand and enforced server-side.

Connected in ten minutes. Useful the same session.

The install is one config entry. Everything after that is a conversation in the client you already use.

• Ten minutes: Add the server and run a read-only call. — Your agent lists the brand, its voice rules and the keyword bank — proof the key and the scope are right.
• First session: Ask it to write one article end to end. — A draft in the brand voice with structure, internal links and images, sitting in the queue for you to read.
• Day two: Wire it into whatever you already run. — A cron, a CI step, or a slash command in your own repo — the platform does not care who is calling.
• Day five: Point a second brand at the same setup. — A separate key, a separate voice, the same tools. This is how an agency of one runs ten clients.

When this is the wrong tool.

• You want a point-and-click product and never want to see a config file. The dashboard is complete on its own — but then the MCP surface is not what you are paying for, and the operator page is the better read.
• You need the content pipeline to run fully inside your own infrastructure. This is a hosted platform with an API, not a library you deploy.
• You want an open-ended agent that publishes without a human in the loop. Autonomy is a dial on the brand, enforced on the server, and it starts closed.
• You are building a competing writing product and want the model prompts. The generation contract is the platform’s, and it is not exposed.