PowerScroll

What PowerScroll MCP server does

PowerScroll MCP server connects AI agents to an open PowerScroll notebook running in a browser. PowerScroll stores the editor and notebook data together in one editable HTML file, so the bridge operates on the same workspace that a person sees and edits. The notebook supports sections, pages, independent named scrolls, text, images, drawings, shapes, and diagram content.

Agents can inspect notebook content, add or reposition blocks, create or manage scrolls, insert images, generate native editable diagrams, and save changes. Diagram sources can include PlantUML, Mermaid, SVG, or draw.io content. Human and agent changes use the same in-memory state, undo history, and save path rather than maintaining a separate agent database.

How it works

The bridge runs locally and is disabled until the user explicitly enables it in the notebook. After the local package is started, open a notebook and select Settings → Agent bridge → Let a local agent write into this notebook. The bridge then accepts local agent access through a loopback WebSocket connection.

PowerScroll supports local multi-agent use with a single-writer lease. Reads can continue without being blocked, while the lease coordinates writing access. This arrangement is intended for agents working against the currently open notebook, not for a hosted or remotely shared notebook service.

Notebook data remains in the HTML file. The hosted demo does not upload notebook content, and normal notebook editing does not require an account or a PowerScroll server. Update checks and optional extension installation can contact GitHub, but those are separate from the local bridge connection.

Setup and configuration

Install and run the published bridge package with:

npx -y powerscroll-mcp

The package is published on npm as powerscroll-mcp; npm is the documented canonical installation source. The latest GitHub release also provides a package tarball as a fallback. After starting the package, open the target notebook and enable the agent bridge from Settings. No environment variables or account configuration are documented in the supplied material.

A notebook can be opened directly in a browser from a downloaded PowerScroll.html file or from the hosted application. Pressing Ctrl+S saves the application and notebook together. Chrome and Edge support direct disk save and autosave, while other situations can use the browser download fallback.

Tools and capabilities

The bridge supports operations for:

• Reading pages and working with notebook structure.
• Adding and moving blocks on the canvas.
• Creating and managing named, resizable scrolls.
• Inserting images.
• Creating editable diagrams from supported diagram formats.
• Saving the current notebook.

The underlying notebook also provides markdown, tables, checkboxes, links, syntax blocks, KaTeX mathematics, freehand drawing, geometric shapes, search, replace, outline navigation, pan, zoom, lasso selection, alignment guides, and undo/redo. These editor features provide the workspace context for agent changes, although the supplied material does not define a separate MCP tool for every feature.

Limitations and notes

The bridge is local-only and requires the user to enable it before an agent can write. PowerScroll is an executable HTML document, so notebooks should be opened only from trusted sources. The README describes a local bridge and local notebook workflow; it does not document remote hosting, authentication, environment variables, or a cloud synchronization service.

Existing PowerNote notebooks can be opened directly. For older notebooks that cannot update themselves, open them with the latest PowerScroll HTML file and save them again to write the current editor around the existing content.