# pmx-canvas [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/pskoett/pmx-canvas  
**GitHub Stars:** 30  
**npm Downloads (last month):** 2241  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/pmx-canvas

## Description
PMX Canvas is a shared board where you and your AI agents collect and present the data and evidence behind your work. Curate that material into context for your agent's next step, then bring the results back onto the board to review and build on. You both add and edit material on the same board, in real time. Pin context to focus input and output.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `bunx` (confidence: high):

```json
"mcpServers": {
  "pmx-canvas": {
    "command": "bunx",
    "args": ["pmx-canvas"]
  }
}
```

## Documentation

## What pmx-canvas does

pmx-canvas provides a shared, local workspace where people and AI agents organize the material behind a task. A board can contain notes, files, web pages, images, charts, code diffs, diagrams, and richer interactive views. Cards can be arranged on an infinite canvas, grouped, connected with labeled links, and placed in folders for later retrieval.

The MCP connection lets an agent work with that board instead of relying only on the current conversation. You can pin cards or entire boards to select the context for an agent's next step. The agent can then return findings, summaries, task updates, or other material to the board for review and further editing.

## How it works

Boards are stored locally in `.pmx-canvas/canvas.db` within the project folder. Reopening the canvas from that same folder preserves the board across sessions. The application runs locally, and connected agents share the open board rather than using built-in multi-user accounts.

The canvas supports evidence-oriented workflows. The board can also be presented as a guided tour or exported as a self-contained HTML file. Undo, snapshots, and backup commands provide ways to manage board history and copies.

## Setup and configuration

The recommended setup uses Node.js with npm and the PMX Canvas skill:

```bash
npx skills add pskoett/pmx-canvas/skills/pmx-canvas
```

The skill guides the agent through installing the application, connecting to it, starting the canvas, and handling required permissions.

For manual use, install Bun 1.4.4 or newer and run the application from the folder where boards should be stored:

```bash
bunx pmx-canvas
```

Use `bunx pmx-canvas --demo` in an empty folder to open a sample board. A global installation is also available with `bun add -g pmx-canvas`, followed by `pmx-canvas`.

An MCP client can launch the server with `bunx pmx-canvas --mcp`:

```json
{
  "mcpServers": {
    "canvas": {
      "command": "bunx",
      "args": ["pmx-canvas", "--mcp"]
    }
  }
}
```

Start the agent in the same project folder as the canvas, then approve or enable the connection in the client. A plugin is another option for clients supporting the Agent Plugins 1.0 standard and MCP stdio connections.

## Tools and capabilities

- Read material selected through pinned cards and boards.
- Add findings and other agent-generated material to a board.
- Edit board content and update task-related material.
- Organize cards with groups, folders, links, annotations, search, and a board map.
- Build dashboards, comparisons, and interactive views on the canvas.
- Review agent changes and see what an agent has read or added.
- Present boards as tours or export them as standalone HTML.

## Limitations and notes

PMX Canvas has no built-in multi-user accounts. Agents connected to the same open board share its content. Material requested from an agent may pass through that agent's tools and model provider; web content and hosted applications may also involve their external services.

Autosave writes to the project-local database but is not a backup. Use the documented backup and restore commands when a separate copy is needed. Client setup varies, and a client-specific plugin or direct MCP configuration may be required.

_Full upstream README: https://allmcps.com/mcp/pmx-canvas/readme_