Skip to main content

Build with AI

The studio is also a Model Context Protocol (MCP) server. An AI agent (Claude Code, Claude Desktop, Cursor, or anything that speaks MCP) can work in your workspace with no UI:

  • Draft and revise: one piece from a prompt or from an AI visibility item, in your voice, as a blog post, a LinkedIn post, an X post or thread, a newsletter or a custom kind; rewrite it from an instruction; check its claims; restore a version; export it as Markdown, Word or PDF.
  • Teach the brand: the voice, the words you never say, the standing rules, the rule per kind, the docs, the checks and what they learned.
  • Read AI visibility: visibility, share of voice and position, each question's answers, the pages cited, measured changes, the next steps and your pages; managers edit the questions.
  • Prepare for your website: the preview of a draft for a CMS destination, a proposed change to one of your pages, and where a piece stands in your CMS.
  • Run your repo's copy through the brand: review and rewrite the copy in your codebase without creating drafts.

Two things stay a person's press in the app, from the app, the MCP server and Ask alike: saving a draft or a page change in your CMS, and publishing it. An agent prepares and reads; a manager presses. Nothing over MCP posts to a social network, because nothing in Viewfield does.

A few things are the app's alone, by design: live web research before writing (Create and Ask Viewfield), connecting a CMS or Google, exporting to Google Docs, members and roles, API keys and billing.

Two ways to start​

1. Install the server in your client​

npx @viewfieldai/mcp install

It prompts for your API key, detects your client and wires Viewfield in. See Install the MCP and Authentication.

2. Call it from your own code​

The server is reachable at https://viewfield.ai/api/mcp over the Streamable HTTP transport, stateless, answering POST only, with your API key as a Bearer token. Any MCP client library works; the key scopes every call to one workspace.

The server is self-describing​

There is no per-tool reference to memorize. Your client reads each tool's name, description and input schema over the protocol at connect time. Ask your agent to "list the Viewfield tools". MCP overview says what the tools cover, area by area.