> ## Documentation Index > Fetch the complete documentation index at: https://docs.getprova.com/llms.txt > Use this file to discover all available pages before exploring further. # MCP Access: Connecting to Claude and ChatGPT > Learn how MCP access lets Claude, ChatGPT, and other compatible assistants work directly with live Provalytics data inside your existing workflow. ## Overview MCP access is one of the most workflow-changing capabilities in Provalytics. It gives compatible AI assistants direct, read-only access to live Provalytics data through the Model Context Protocol, so teams can ask questions in conversation instead of exporting files, building one-off spreadsheets, or writing ad hoc queries. This is where many marketing and agency teams save hours each day. ## Why this matters With MCP access, an assistant can answer questions like: * What are our top channels by incrementality this month? * What does the optimizer recommend next? * Which campaigns are underperforming relative to spend? * How well did the latest model fit actual results? That means the workflow becomes: * ask * inspect * decide instead of: * export * clean * reconcile * summarize * then decide ## What MCP access unlocks MCP access is especially useful for: * agency teams building faster client readouts * marketers pressure-testing results before meetings * planning teams comparing options without waiting for a custom export * leadership support when a fast answer is needed from live data ## How access works Provalytics provides a client-scoped API key that authenticates requests to the Prova MCP server: * `https://mcp.getprova.com` Each key is scoped to a specific client’s data. ## What the MCP access page looks like The MCP Server Access page gives you everything needed to set up a client-scoped connection: * client workspace selector * key generation form * active keys list * ready-to-use connection instructions for Claude Desktop * CLI example for Claude Code * raw API example for direct MCP requests MCP Server Access overview ## Before you connect Before connecting Provalytics to Claude or ChatGPT, make sure you have: * access to the `MCP Server Access` page in Provalytics * a client workspace selected * a user-specific API key generated for your account * a supported AI workspace that allows MCP or custom connector setup Each key is unique to the user and scoped only to that client workspace. ## Setup guides Choose the guide that matches where you want to work: * [Connect Provalytics MCP to Claude](/integrations/connect-provalytics-mcp-to-claude) * [Connect Provalytics MCP to ChatGPT](/integrations/connect-provalytics-mcp-to-chatgpt) ## What data the assistant can access The internal MCP feature spec currently exposes tools for questions about: * campaign performance * incrementality * recommendations * marginal response * categories / funnel data * model predictions * model statistics * CPM * days to conversion * methodology ## Security notes * keys are client-scoped * keys are read-only * keys should be treated like credentials and not shared between users * keys should be revoked when no longer needed * usage can be monitored through the MCP usage view ## Why teams love this MCP access makes Provalytics part of the team’s natural working environment. Instead of opening a dashboard first, exporting data, and translating it manually, the team can work directly through Claude or ChatGPT and stay in the flow of the conversation. That is where the time savings compound.