Connect your agent
A coding agent can work on your Vizitka site through MCP. You approve the connection in your browser, and the agent receives access only to sites in your account.
Claude Code
Section titled “Claude Code”Run this command in a terminal:
claude mcp add --transport http vizitka https://ai.vizitka.app/mcpThe command saves the server address. In Claude Code, authenticate the Vizitka server when prompted. If needed, run:
claude mcp login vizitkaYour browser opens auth.vizitka.app. Sign in with Instagram, check the agent
name and account, choose the access it should have, and approve the connection.
The agent connects to ai.vizitka.app; the sign-in and approval screen lives
on the separate auth address.
Cursor
Section titled “Cursor”Open Cursor’s MCP settings, or add this remote server to your personal
mcp.json:
{ "mcpServers": { "vizitka": { "url": "https://ai.vizitka.app/mcp" } }}Enable the server and follow Cursor’s OAuth prompt in your browser.
ChatGPT
Section titled “ChatGPT”Custom MCP apps require developer mode in ChatGPT. In Settings > Apps, open
the advanced settings and enable developer mode if your plan and workspace
allow it. Choose Create, enter https://ai.vizitka.app/mcp as the remote
server, select OAuth, scan the tools, and complete the browser prompt.
Availability and write access depend on your ChatGPT plan and workspace settings.
Vizitka registers compatible clients when they first connect, so there is no client ID, secret, or token for you to copy. If a client asks you to paste a token, do not. Use its normal Connect or Sign in action instead.
Choose access
Section titled “Choose access”Read access lets an agent inspect the site, its pages, files, versions, languages, and publishing state. Write access also lets it create drafts, upload files, change settings, and publish. A connection never reaches a site outside your account.
Disconnect an agent
Section titled “Disconnect an agent”Open Dashboard > Account. Under Coding agents, find the connection and select Disconnect. Its current access stops immediately. The client must ask for permission again before it can return.