Connect Harvv to Copilot with one config block. Copilot reads your live friction findings (dead clicks, rage clicks, broken forms) right inside your editor, then helps you fix them.
The token scopes the connection to your account. Every request Copilot makes only ever sees your sites. The raw token is shown once, so copy it when it appears.
In VS Code, create or open .vscode/mcp.json in your project (or run the command "MCP: Add Server" from the command palette). Paste this. Copilot will prompt you for the token the first time it connects, so it never has to live in committed config.
Using the Copilot CLI or another client instead? Same server, same token. Point any MCP client at https://harvv.com/mcp with header Authorization: Bearer YOUR_TOKEN.
Open Copilot in agent mode, enable the Harvv tools when prompted, and ask away. Copilot picks the right Harvv tool, runs it, and explains the result next to your code.
The Harvv MCP returns aggregated findings only: top friction patterns ranked by sessions affected, counts, and which pages are worst hit. It never returns raw event streams, individual user sessions, or any PII. The connection is read-mostly. The only tool that writes anything is the pixel installer. Nothing in the tool set can delete or change your analytics data.
Plain-language prompts. Copilot maps each to the right Harvv tool.
Harvv runs a Model Context Protocol (MCP) server scoped to your account. MCP is the open standard that lets AI assistants like GitHub Copilot call external tools. Once you add the Harvv MCP to Copilot, you can ask Copilot questions like "what friction is hurting my site this week" and it pulls the answer straight from your Harvv data, without you leaving the editor.
GitHub Copilot in VS Code (agent mode) and the GitHub Copilot CLI both support custom MCP servers. The same config also works in Cursor, Claude Desktop, Windsurf, Zed, and any other MCP-compatible client. It is one standard config block; only the file you paste it into changes.
No. The MCP returns aggregated findings only: top friction patterns ranked by sessions affected, counts, and which pages are worst hit. It never exposes raw event streams, individual user sessions, or any personally identifiable information. The same PII scrubbing that runs in the pixel applies before anything reaches the MCP. Copilot can interact with your findings, but it cannot pull your logs.
The Harvv tools are read-mostly. Copilot can read your findings, list your sites, deep-link to your dashboard, and check whether your pixel is firing. The only action that writes anything is the pixel install helper, which adds the Harvv snippet to a project you point it at. There is nothing destructive in the tool set, and Harvv never ships code changes to your app on its own.
Sign in and click Generate token on this page. The token is scoped to your account only: every tool call filters to your sites and no one else's. The raw token is shown once. If it ever leaks, revoke it from your dashboard and generate a new one. Treat it like a password and keep it out of public repos (use an input variable in your MCP config so it is not committed).
Try: "Use Harvv to show the top friction on my site this week." "Which page has the most rage clicks?" "Is my Harvv pixel installed on staging?" "Open my Harvv dashboard for this site." Copilot picks the right Harvv tool, runs it, and explains the result in plain language next to your code.