Getting started
Overview
What this is
The Model Context Protocol is a standard way to hand an AI assistant a set of tools. The Vidrys MCP server is one of those: point Claude, ChatGPT, Cursor or Codex at it and the assistant can read your workspace — which questions you win, who beats you where, what the answers cite, what to write next — and, on Scale and Custom, act on it.
https://api.vidrys.com/mcp| Detail | |
|---|---|
| Transport | Streamable HTTP, stateless, JSON responses |
| Authentication | An API key, or a sign-in connection where your workspace has one |
| Tools | 24 read, 5 write |
| Data | The same parity DTOs as the REST API, returned as structured output |
What you can ask
Tools are useful in proportion to the questions people actually have. These all resolve to one or two calls:
- “How did our AI visibility move this week, and which engine got worse?”
- “Which buyer questions are we losing on ChatGPT, and who wins them instead?”
- “Why did we lose that Perplexity answer — give me the reasons and the actions.”
- “Which third-party sites do the answers we lose keep citing?”
- “What should we publish next, and what does the brief say?”
- “Did the page we published last month actually change anything?”
list_projects. Every other tool takes a project_id, and an assistant that guesses one gets a not_found.The tools
Each tool mirrors an endpoint in the REST reference and returns exactly the same fields. The sidebar lists all 29, grouped the way the dashboard is: visibility, prompts, results, competitors, gaps and citations, content, the queue, outcomes and the site.
Write tools — adding prompts, moving content opportunities and queue tasks, generating a draft — are marked write and need a read-and-write credential on Scale or Custom. Generating a draft spends a content credit and asks for explicit confirmation before it does.
Who can use it
MCP access is included on Grow, Scale and Custom, and the workspace has to be approved. A credential that is genuine but not entitled — a workspace that moved down to Track, say — still connects, and every tool then answers with the reason rather than a bare 401. That is deliberate: a client told “unauthorized” tends to loop through sign-in forever, while a client told plan_required can tell the person what to do.