Skip to main content
Glama

PureRef MCP

An MCP server that lets an AI agent (Claude Code, Claude Desktop, or any MCP client) see and build PureRef reference boards.

It drives PureRef 2.1+'s official command-line interface. Each call runs PureRef as a separate headless process, so an open PureRef window is never touched.

What it can do

Tool

What it does

pureref_list_scenes

Lists the .pur boards in a folder

pureref_scene_info

Shows where each image on a board came from (local file or web URL) without launching PureRef

pureref_view_scene

Renders the board with a scene-coordinate grid, optionally zoomed to a region

pureref_view_images

Shows a numbered contact sheet of the board's images

pureref_export_images

Exports full-resolution copies of the images

pureref_render_png

Renders the board to a PNG

pureref_add_items

Places images (paths or URLs) and text labels at exact positions

pureref_add_image_grid

Places a grid of images

pureref_build_board

Lays out a titled board from sections of images in one go

pureref_open

Opens a board in the PureRef app

pureref_list_backups

Lists the automatic backups of a board

How it stays safe

  • Backups: before a board is written, the original is backed up. The last 20 backups per board are kept in %LOCALAPPDATA%\pureref-mcp\backups.

  • Open boards: writing a board that's open in the PureRef window is refused unless allow_open=true.

  • Atomic saves: saves go to a temp file first, then replace the original.

Limits (from PureRef's CLI)

  • It can't move, resize, delete or relabel items that are already on a board. It can only add new ones.

  • Text labels are added as images, not native PureRef notes.

  • Reopen the board in PureRef to see the changes.

Related MCP server: gimp-agent-mcp

Requirements

  • Windows. The CLI and exe lookup are only tested there.

  • PureRef 2.1 or newer.

  • Python 3.11+.

Setup

  1. Install the Python packages:

    pip install -r requirements.txt
  2. Register the server with your MCP client. For Claude Code:

    claude mcp add pureref -s user -e PUREREF_MCP_ROOT="C:\path\to\your\boards" -- python "C:\path\to\PureRef-MCP\server.py"

    For Claude Desktop, or any client that uses an mcpServers JSON config:

    {
      "mcpServers": {
        "pureref": {
          "command": "python",
          "args": ["C:\\path\\to\\PureRef-MCP\\server.py"],
          "env": { "PUREREF_MCP_ROOT": "C:\\path\\to\\your\\boards" }
        }
      }
    }

Settings (environment variables)

Variable

Default

What it's for

PUREREF_MCP_ROOT

~/Documents/PureRef

The folder that relative board names and pureref_list_scenes use

PUREREF_EXE

Looks in Program Files

Path to PureRef.exe, if it's installed somewhere else

PUREREF_MCP_CACHE

%LOCALAPPDATA%\pureref-mcp

Where previews, downloads and backups go

Coordinates

  • Positions are in PureRef scene units: 1 unit is 1 image pixel at 100% zoom.

  • x grows to the right and y grows downward.

  • Image positions are centers unless you pass anchor="top_left".

  • pureref_view_scene draws the grid (cyan x= lines, yellow y= lines), so the agent can read positions off the preview.

License

MIT, see LICENSE. PureRef is a product of Idyllic Pixel AB. This project isn't affiliated with or endorsed by them.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      B
      quality
      B
      maintenance
      Enables AI agents to operate GIMP 3 end-to-end: open and inspect images, call every PDB procedure, apply GEGL filters destructively or as layer effects, measure pixels, render before/after/diff comparisons, cut out subjects with AI segmentation, and run multi-step recipes across folders.
      32
      76 PyPI
      5
      Apache 2.0
    • F
      license
      Not graded
      quality
      A
      maintenance
      Enables AI agents to control a live Blender session, inspect and modify scenes, manage PolyHaven assets, render images, and route prompts through configurable local or OpenAI-compatible LLM backends.
      -