project libraryactive project: launchpad
Bootstrap visual design with desktop + MCP
Bootstrap visual design fast.
Strapping AI Canvas is the fastest way to go from a blank surface to a structured visual direction. Open the desktop app, point an MCP client at the same project, and rough in scenes, layout, and copy before polishing by hand.
Fast first pass, local runtime, explicit boundaries, one shared document core.
mcp session127.0.0.1
endpoint: http://localhost:9311/mcp tool: inspect_project tool: apply_commands commands: - create_scene id=scene_launch - create_node id=rect_shell kind=rectangle - create_node id=text_headline kind=text - inspect_scenes mode: read_write
Bootstrap screens fast
Desktop + MCP parity
Scene-first structure
Local projects
Deterministic edits
Why Strapping
The quickest path from blank canvas to structured direction.
Strapping AI Canvas is built for the messy first pass: not final polish, but getting a usable visual direction into view quickly. Speed comes from one local project model shared by the desktop UI and MCP clients.
Fast First Pass
Start with a structured first pass instead of a blank page. Rough in scenes, layout blocks, copy, and system primitives quickly, then refine from a visible direction.
One Runtime
The editor UI and the local MCP bridge use the same document schema, command system, and semantic logic. Agent edits do not disappear into a second automation layer.
Structured Mockups
Scenes, variables, styles, bindings, and provenance give the model something better than pixels alone, so bootstrap edits stay inspectable and usable.
Workflow
Open locally. Strap in an agent. Tighten in the same canvas.
The loop is short: open a local project, connect MCP, generate a structured first pass, and refine the same live result in the editor.
Open a local workspace
Create or open a local project in the desktop app and establish the workspace you want to push forward.
Bootstrap with an agent
Connect Claude, Codex, or Gemini through the local MCP endpoint and generate first-pass structure, copy, and layout against the same active project session.
Refine in the canvas
Use the canvas, layers, and inspector to tighten what the agent started without exporting, translating, or losing state across tools.
workflow scenecanvas + layers + inspector
One workspace carries the whole loop: bootstrap the structure fast, inspect the result, and keep tightening in the same scene-first editor.
speed leverOne active project session
best fitFirst-pass screens and system scaffolds
shared pathUI, MCP, undo/redo, autosave
MCP
The speed story is one runtime, not automation glue.
Strapping AI Canvas exposes a local MCP surface over the same command/query/document core the desktop editor uses. That is why the bootstrap pass stays fast, inspectable, and close to the real product state.
available toolsfirst pass
list_projects
create_project
open_project
inspect_project
inspect_tree
inspect_node
inspect_scenes
inspect_design_system
apply_commands
installationendpoint + client buttons
Use the full localhost endpoint, including the /mcp path, and keep the editor window open for write-capable actions.
runtime modescapability boundary
read_write
Available while the editor window is open and the measurement surface exists.
read_only
Inspection remains available after close. Mutation fails fast with measurement_surface_unavailable.
Trust
Speed without ambiguity beats vague AI promises.
Fast only works if the product boundary stays obvious. This is a local-first desktop tool with a visible runtime contract, and the marketing story should match that contract.
Projects, metadata, and assets stay local.
No mandatory backend or account setup.
Core editing workflows are offline-capable.
MCP is localhost-only and user-visible.
UI and MCP share one document model.
Write boundaries stay explicit and deterministic.
Not a browser-first editorNot cloud collaborationNot realtime multi-userNot multi-document per project in v1
Scene-first structure
Structured design data gives the model something solid to build on.
Scenes, variables, styles, bindings, and provenance are first-class project data. That gives both the editor and an agent a stable surface for inspection and mutation.
document treescene-first
- project: launchpad
- document: current_document_json
- scene: scene_launch
- frame: scene_launch
- rectangle: rect_shell
- text: text_headline
- style: hero_surface
- variable: color.neutral.950
design systemvariables + styles
- variables: color.neutral.950, type.display.lg
- styles: hero_surface, body_copy
- bindings: text_headline -> hero_copy
- provenance: command batch / revision 12
Current status: sharp first-pass workflow, active product development.
Strapping AI Canvas is still under active development, but the direction is clear: get from zero to usable visual design faster by keeping desktop editing and agent actions in one local workspace.
FAQ
What the site should answer directly.
Short answers to the runtime and scope questions that matter when evaluating the product.
Why desktop instead of browser-first?
Strapping AI Canvas is built as a standalone Electron app so the editor, local persistence, and MCP runtime can share one local-first document model without a mandatory backend.
What does bootstrapping visual design mean here?
It means getting from a blank canvas to a credible visual direction quickly: scenes, layout blocks, headline copy, and reusable system primitives before detailed polish.
Does Strapping AI Canvas require the cloud?
No. Core editing workflows are designed to work locally without a running server or cloud control plane.
What happens when the editor window closes?
Closing the editor window does not quit the app. The app stays resident in the tray, and MCP inspection remains available against the active project session.
Can agents still inspect projects after close?
Yes. Inspection remains available after the window closes until the user explicitly quits the tray app.
Can agents mutate while the window is closed?
No. In v1, mutations fail fast with measurement_surface_unavailable when the editor window is closed because the browser-backed measurement surface is unavailable.
Does a project contain multiple documents?
No. In v1, each project contains exactly one document. Project targeting is the primary MCP target surface.
Download
Bootstrap your next interface direction in one local workspace.
Download the latest desktop build, inspect the docs, or review the source. The current release is experimental, but the value is immediate: faster first-pass visual design with a local editor and MCP in the same loop.