PrixmaViz

▶ Watch the 30-second demo in HD

AI-native diagram tool. 28+ rendering engines (Mermaid, PlantUML, D2, Vega-Lite, TikZ, Graphviz, WaveDrom, Bytefield, …) wrapped behind an MCP server, with annotations, infinite-canvas multi-tile workspace, and a marketing-surface workspace UI. Cycle 4 ships as a hosted service with optional self-host.

Try the hosted version

Visit https://prixmaviz.ailuxis.com — your workspace is created automatically and the URL is bookmarkable. Lose the URL, lose the workspace, so save it somewhere.

Install

Supported platforms: macOS (Apple Silicon + Intel), Linux (x64 + arm64), Windows (x64). The plugin install itself is tiny (~3 KB launcher script). On your first diagram render, the launcher downloads the matching pre-built shim binary (~60–115 MB depending on platform) from this repo's GitHub Releases and caches it at ~/.cache/prixmaviz/bin/ (POSIX) or %LOCALAPPDATA%\prixmaviz\bin\ (Windows). Subsequent calls hit the cache — zero network.

Claude Code

claude plugins marketplace add MichaelDanCurtis/PrixmaViz
claude plugins install prixmaviz@prixmaviz

Codex CLI

codex plugin marketplace add MichaelDanCurtis/PrixmaViz

Then enable the plugin in ~/.codex/config.toml:

[plugins."prixmaviz@prixmaviz"]
enabled = true

Restart Codex. Your next prompt that mentions diagrams will route through PrixmaViz.

VS Code (any MCP-capable extension)

PrixmaViz is a standard stdio MCP server, so any VS Code extension that supports MCP servers can use it — no PrixmaViz-specific extension is needed. The configuration shape is the same for every host; just paste the snippet below into your extension's MCP config and update the absolute path to wherever the shim binary lives on your machine (after a first CC/Codex install it's cached at ~/.cache/prixmaviz/bin/prixmaviz-mcp-0.9.0-<platform> on POSIX or %LOCALAPPDATA%\prixmaviz\bin\ on Windows; on a fresh machine, download the matching prixmaviz-mcp-<platform> from the v0.9.0 release).

Universal snippet:

{
  "mcpServers": {
    "prixmaviz": {
      "command": "/absolute/path/to/prixmaviz-mcp",
      "env": {
        "PRIXMAVIZ_REMOTE_URL": "https://prixmaviz.ailuxis.com"
      }
    }
  }
}

Where to put it:

• GitHub Copilot Chat (VS Code, recent versions with MCP support) → workspace .vscode/mcp.json or user settings.json under the MCP servers key. See the Copilot Chat MCP docs for the exact path.
• Cline → command palette → Cline: Open MCP Settings (writes to ~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json on macOS).
• Continue → edit ~/.continue/config.json and add the mcpServers block under experimental (or top-level — depends on Continue version; see their MCP docs).
• Roo Code, Cursor (the IDE), and other MCP-aware tools → each has its own settings UI, but the JSON shape above is what they all expect.

Once configured, restart the extension. Asking the AI for a diagram should now route through PrixmaViz the same way Claude Code and Codex do.

Both: configuration

The plugin uses https://prixmaviz.ailuxis.com by default. To point at your own self-hosted instance, set PRIXMAVIZ_REMOTE_URL in the cached .mcp.json (~/.claude/plugins/cache/.../<version>/.mcp.json or ~/.codex/.tmp/marketplaces/prixmaviz/plugin/.mcp.json). To run a locally-built binary instead of the released one, set PRIXMAVIZ_MCP_BIN to its absolute path.

The AI will now use PrixmaViz whenever you ask for diagrams. Try:

Show me the TCP handshake.

CLI

@prixmaviz/cli is the scriptable companion to the MCP plugin — push DSL files to your workspace, pull rendered diagrams to disk, list the library, and round-trip whole workspaces as .pviz bundles. Useful for CI pipelines, doc-site builds, and shell-driven workflows where you don't want an AI in the loop.

npm install -g @prixmaviz/cli

prixmaviz login                          # one-time: server URL + workspace token
prixmaviz push docs/diagram.mmd          # render + save (engine auto-detected from extension)
prixmaviz pull service-topology          # download the rendered SVG
prixmaviz list --tag infra               # browse the library
prixmaviz export-workspace --out ./bak   # snapshot the whole workspace to a .pviz bundle

Standalone binaries (no Node required) are also published to each cli-v* GitHub Release for darwin-arm64, darwin-x64, linux-arm64, linux-x64, and windows-x64.

See packages/cli/README.md for the full command reference, config-file paths, and engine auto-detection table.

Self-host with Docker

git clone https://github.com/MichaelDanCurtis/PrixmaViz
cd PrixmaViz
cp .env.example .env
docker compose up -d

Open http://localhost:5180 — your workspace is ready. All 6 containers (server, postgres, kroki orchestrator, mermaid, bpmn, excalidraw sidecars) come up automatically. PlantUML is built into the orchestrator.

For production, set:

• PRIXMAVIZ_PUBLIC_URL — the canonical URL diagrams share (e.g., https://prixmaviz.yourcorp.com)
• DATABASE_URL — your external Postgres (omit to use the bundled one)
• TLS terminator in front (nginx, Caddy, Cloudflare, etc.)

Workspace lifecycle

Anonymous workspaces expire after 1 hour of inactivity. Activity means any authenticated /api/* request — keeping a browser tab open with the workspace loaded is sufficient.

Pin a workspace indefinitely by toggling at least one diagram to public (🌐 icon on the tile header). Workspaces with any public diagram are exempt from TTL.

To change the policy, set in your .env:

• PRIXMAVIZ_WORKSPACE_TTL_MINUTES=60 — how long before idle workspaces expire (set 0 to disable)
• PRIXMAVIZ_REAP_INTERVAL_MINUTES=5 — how often the reaper runs

Architecture

• Bun + TypeScript server: HTTP + WebSocket + 11-tool MCP dispatch, Postgres persistence, Bearer-token workspace auth
• React + Zustand web client: infinite canvas with tiles, annotation overlay, public-view toggle, settings panel
• @prixmaviz/shim binary: thin MCP stdio→HTTP forwarder distributed via the CC plugin (5 platform builds: darwin-arm64/x64, linux-arm64/x64, windows-x64)
• Postgres 16: 3 tables (workspaces, diagrams, annotations) with workspace-scoped queries
• Kroki for rendering (configurable via KROKI_URL)

See docs/superpowers/specs/2026-05-11-prixmaviz-cycle-4-design.md for the design rationale.

Visio (.vsdx) support

PrixmaViz natively renders, imports, and exports Microsoft Visio diagrams:

• Drag-drop a .vsdx onto the canvas to render it (server-side via the prixmaviz-vsdx sidecar, which runs unoserver/LibreOffice).
• AI translation — ask Claude/GPT/etc. "convert this Visio diagram to Mermaid" and the AI calls analyze_vsdx to get structured shape data, then generates DSL with create_diagram. No server-side LLM is used.
• Export any graph diagram as .vsdx via the tile menu. Mermaid/D2/Graphviz emit Visio-editable shapes from a ~35-shape stencil. Other engines produce an image-embed .vsdx.

Self-host requires the prixmaviz-vsdx sidecar from docker-compose.yaml. Default upload cap is 5MB (VSDX_MAX_BYTES).

Project structure

packages/
  shared/       # TypeScript types shared between server, web, and shim
  server/       # Bun HTTP+WS+MCP server, Postgres repos, hit-testers
  web/          # React webview — InfiniteCanvas, Tile, public view, settings
  shim/         # MCP stdio→HTTP forwarder (compiled to native binary)
plugin/         # Claude Code + Codex CLI plugin (.claude-plugin/, .mcp.json, bin/)
docs/
  marketing/    # Hero video (HyperFrames source) + GIF/poster
  superpowers/
    specs/      # Cycle design specs
    plans/      # Cycle implementation plans

Cycle history

• Cycle 1 — Foundation: Bun server + Tauri shell + 6-tool MCP + Mermaid IR + .pviz persistence
• Cycle 2.plus — Annotations + multi-canvas + initial install path (10 MCP tools)
• Cycle 3 — Real Claude Code plugin: skills, hooks, slash command, 14 MCP tools
• Cycle 4 — Service-first architecture: docker-compose stack, Postgres multi-tenant, thin shim binary, marketing-surface UI

See docs/superpowers/specs/ and docs/superpowers/plans/ for design + implementation history.

License

MIT — see LICENSE.

PrixmaViz uses Kroki (MIT) for rendering.