A diagram toolkit
Create, arrange, compare and export diagrams in JavaScript. The same core powers the Filemigo canvas.
FILEMIGO FOR DEVELOPERS v0.4.0
Your ideas.
Your tools.
One diagram language.
Build with our diagram toolkit. Let your AI create, check and export editable diagrams through a local MCP server.
● ● ●filemigo / local
# From an idea to an editable file you Draw an order approval flow. assistant → filemigo instructions() create_diagram({ … }) validate_diagram({ … }) export_diagram({ format: "svg" }) JSON in. A diagram you can keep.
Illustrative tool sequence. Try the actual core below.
54 vector shapes16 diagram templates18 MCP tools0 API keys required
THREE WAYS TO BUILD
Use the part you need.
A useful starting point
Discover flowchart, UML, ER and network shapes. Start from 16 editable diagram templates.
Give your AI the tools
Connect a compatible local MCP client. It can call the toolkit and return a project you can edit here.
TRY THE REAL CORE
Change the layout. Keep the logic.
This example runs locally with the downloadable SDK. No AI service is called.
UP AND RUNNING
Developer SDK.
Optional source setup.
Install Node.js 22+, download the repository and install its locked dependencies. The first installation needs internet access.
The package is available on GitHub; it is not published on npm. Your AI client’s model costs and data handling are separate.
Read the full guide ↗git clone https://github.com/latecafe-tech/filemigo-canvas.git cd filemigo-canvas npm ci --ignore-scripts npm test
2. Configure your MCP client
Replace the path with your extracted folder. On macOS/Linux use /absolute/path/filemigo-canvas/src/mcp.mjs. This is a common configuration format; exact fields depend on your client. It must support local stdio servers.
FILEMIGO DESKTOP / v0.4
Your canvas. On your computer. With AI.
Download the Windows app for offline drawing and local MCP. The website remains an installation-free editor; live MCP runs in Desktop.
Download for Windows · 64-bit ZIP ↓ Setup guide ↗
Extract and open
Extract the complete ZIP into a permanent folder and run Filemigo.exe. No Node.js installation. Open and save local projects; export SVG, PNG and PDF.
Copy your configuration
Open MCP setup in Desktop. Copy its JSON into a local stdio MCP client, restart the client, then enable AI in the app. Ask for filemigo_live_start.
See changes as they happen
AI edits the open desktop canvas. Revision checks protect manual work; locked shapes stay locked. Pause, disconnect or undo. You save the files.
First Windows release: portable and unsigned; no automatic updater. macOS/Linux packages are not included yet. The canvas works offline; your AI client may need internet and its own subscription. The enabled canvas is shared with that client.
MCP APPS / v0.3
See it. Review it. Keep editing.
Interactive chat preview
Call filemigo_show_diagram in an MCP Apps-compatible local client. Zoom, edit text, undo, arrange and download inside the preview. Ordinary MCP clients still receive JSON.
Review before applying
filemigo_preview_changes reports additions, removals and meaning changes without mutating your canvas. Read the current desktop revision before applying an approved change. Stale writes are rejected.
Editable file exchange
Import supported draw.io XML (including compressed pages). Export basic shapes and relationships to draw.io. Editable SVG embeds the complete Filemigo project. Multiple layers, nested groups, specialist stencils and sequence scopes require native formats.
This is a local stdio MCP App, not a hosted remote MCP endpoint or a store listing. Preview editing does not automatically update a paired canvas. Use Open in Filemigo or save the project.
Know what this release does.
The toolkit returns data and SVG. Save native JSON and open it in Filemigo to continue editing. Connect Filemigo Desktop to see AI edits immediately. The website also offers an independent editor.
Layout checks preserve explicit relationships. Validation cannot prove business intent. Review conditions, connection warnings and change reports.
Up to 150 nodes and 200 edges. Supported draw.io XML import/export, with explicit limitations. MCP returns results to your AI client, whose privacy settings apply. Desktop uses an authenticated local pipe; AI access starts disabled. No hosted relay or paid API.
OPEN SOURCE / MIT