VibeFactory MCP Server: Build Dashboards from Claude and Cursor

MCP stands for Model Context Protocol, an open standard for letting AI assistants use external tools and services. An MCP server is the piece a product publishes so that Claude, Cursor and other AI tools can talk to it in a structured way. Claude calls these connections connectors; its directory lists more than 950 of them as of mid-2026. VibeFactory's server is a remote MCP server, which means there is nothing to install: you paste an address, sign in, and you are connected.

Short and specific beats long and vague, exactly as in the web app. These are prompts we used while testing:

• "List my uploaded files and build a financial performance dashboard from the newest one. Include KPIs for net revenue, gross margin and units sold, a monthly trend, and breakdowns by region and product line."
• "What other charts could we add to the sales dashboard?" (the assistant reads the dashboard and answers; no credits spent)
• "Add a monthly margin trend chart to it, using the existing region filter."
• "Change the background from dark to light and keep everything else as it is."
• "Check whether the last build finished and give me the link."

Every edit ends with a short list of the files that changed and how many lines, so you can see at a glance that only what you asked for was touched.

If you came here looking for an Excel MCP server, note the difference: most Excel MCP servers let Claude read and edit cells in a workbook. VibeFactory does something else with the same file. Claude hands your Excel or CSV data to VibeFactory, and what comes back is a deployed, interactive dashboard with its own URL that you can share, not a spreadsheet. So "can Claude create an Excel dashboard?" now has a short answer: yes, from this connector, and the result lives on the web rather than inside Excel.

A dashboard always starts from a data file, and there are two ways to provide one:

• From the web app: upload an Excel, CSV or JSON file on the dashboard page as usual, then ask your assistant to build from it. This is the simplest path from Claude, which cannot send files from your computer to other services.
• From Cursor or any tool that can run commands: ask it to upload the file directly. The server hands back a one-line upload command, the file goes straight to secure storage, and the dashboard can be built in the same conversation.

Each uploaded file is used for one dashboard. Files you upload and never use are cleaned up automatically. Google Sheets dashboards, which fetch the sheet's current data on every load, are created in the web app for now.

We kept the first version deliberately narrow. Through MCP, an AI tool cannot:

• Read or edit the source code of your dashboards. If you want the code, connect a project to GitHub in the web app.
• Delete dashboards, change domains or manage sharing. Those stay in the web app, where the confirmation steps are.
• Change dashboards that were shared with you. Only the owner can send instructions to the agent.
• See the password of a private dashboard. Passwords are only ever shown on the project page.

Connecting uses OAuth, the same sign-in flow you use for Google or GitHub integrations. You never paste an API key into a chat tool, and VibeFactory never sees your Claude or Cursor credentials. On the approval page you see exactly which application is asking, and only known AI clients can be approved.

Every action an AI tool takes on your account is recorded in VibeFactory's audit log with the client that made it. If you stop using a tool, open Settings, AI clients and click Disconnect. Every token issued to that client stops working immediately.

The MCP server is included in every VibeFactory plan at no extra cost. Building and editing dashboards spend the same credits as in the web app, so the pricing page applies unchanged. Reading your dashboards, uploads and balance is free, so you can connect, look around and decide later.