MCP (Model Context Protocol)
MCP (Model Context Protocol) is an open protocol, introduced by Anthropic, that lets AI agents discover and call external tools, read resources, and use prompts through one standard interface.
For dashboards, MCP replaces the brittle approach of letting an agent click through a web editor. An MCP dashboard server exposes real operations as typed tools with defined inputs and outputs, so an agent can list, read, validate, and change dashboards the same way a program calls an API.
How it works in dvt
The dvt MCP server exposes nearly 80 tools that MCP clients such as Claude Code, Claude Desktop, Cursor, VS Code, and Codex CLI can call. They cover the whole lifecycle of an agent-built dashboard: discovering folders, dashboards, skills, and chart types; validating a draft; applying it with preview; rendering and diffing versions; reading and resolving comment threads; sharing; and exporting.
Every one of those tools reads or writes the same dashboard spec a person edits in the browser. There is no separate AI format and no translation layer between what the agent wrote and what renders. The quickstart has the exact connection steps, and the MCP dashboard tool page groups the tools by task.