AI Agents & MCP · 6 min read · 2026-11-06
Building for Humans and AI Agents: Model Context Protocol (@privacyfold/mcp) & Deterministic Schemas
Key Takeaway
How a single JSON tool manifest powers both an accessible human interface and deterministic tool-calling for Claude Desktop, Cursor, and autonomous agents.
PrivacyFold uses a single declarative tool manifest in Git (`content/tools/*.json`) to drive both its human UI controls and its Model Context Protocol (`@privacyfold/mcp`) server.
- **Flip the Page:** Peel the top-right corner of the Settings card on any tool to inspect the exact URL API link, JSON-RPC `tools/call` payload, and structured output.
- **Visible Co-Worker Agent:** Click **Watch an agent do this** to see an agent operate the real UI controls with **Pause** and **Take over** buttons.
Try the Tool Privately
Run the client-side tool explained in this guide with zero server uploads.
Open Tool →