Features · Prove
Connect Claude, Cursor or any MCP client and ask questions in plain language. Every answer comes back with the query that produced it, and the server can only read.
The tools cannot create, modify or delete anything. An agent that can write to your analytics can also corrupt the record you use to check its work.
Each tool returns query, params and rows alongside the answer, so the agent is reasoning over verifiable data rather than a summary it has to trust.
The allow-list is resolved from your key before any tool runs. A site you cannot see and a site that does not exist produce identical errors — no probing for which is which.
The MCP tools call the same metric definitions. There is no second pipeline that can drift from the first.
The evidence
{
"mcpServers": {
"vitrus": {
"command": "npx",
"args": ["-y", "@vitrus/mcp"],
"env": { "VITRUS_API_KEY": "vk_live_…", "VITRUS_URL": "https://app.vitrus.dev" }
}
}
}
Then ask: "which pages is ChatGPT reading on my site, and do those visits convert?"