Panel / HoloViews MCP Integration

Panel and HoloViews are powerful Python visualization libraries but they have no MCP integration today. AI agents can write visualization code but the result never renders inline in the chat. There is no way to create, update, or interact with a live dashboard from within an LLM conversation. This project builds an MCP Apps-compatible server for Panel and HoloViews that changes that. MCP Apps let tools return sandboxed, embedded interfaces that render directly inside MCP clients like VS Code, Claude Desktop, and Cursor, with bidirectional communication between the interface and the host. The goal is to turn Panel and HoloViews into first-class interactive UIs inside any agent conversation. The project will deliver three things: a CSP-compliant sandboxed rendering path for MCP Apps covering the full hvPlot and HoloViews chart surface, a bidirectional postMessage protocol so widgets can communicate with the agent in real time, and cloud-native dataset support for Parquet, Zarr, and the Earthmover marketplace. All work will be delivered as tested, documented PRs to the HoloViz ecosystem.