# keiver/image-tiler-mcp-server [Health: Active]

**Category:** 🎥 Multimedia Process  
**Repository:** https://github.com/keiver/image-tiler-mcp-server  
**GitHub Stars:** 3  
**npm Downloads (last month):** 89  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/keiver-image-tiler-mcp-server

## Description
Full-resolution vision for LLMs. Tiles large images and captures web pages via Chrome CDP so vision models process every detail without downscaling. Generates interactive HTML tile previews. Supports Claude, OpenAI, Gemini presets with per-model token math and entropy-based tile classification.

## Tools
Capabilities this server exposes over MCP:

- **tiler** — Split images into optimally-sized tiles for LLM vision analysis, or capture web page screenshots and tile them.

MANDATORY two-phase workflow — DO NOT skip Phase 1:

  Phase 1 (REQUIRED first): Provide ONLY the image source (filePath, sourceUrl, url, etc).
           DO NOT include preset, tileSize, or outputDir.
           Returns a model comparison table with token estimates and an outputDir.
           You MUST present this table to the user and ask which preset they prefer.
           DO NOT select a preset yourself — the user decides. If you must auto-select, always use the cheapest option.

  Phase 2: Call again with the user's chosen preset + the outputDir from Phase 1.
           Re-include your original image source (filePath, sourceUrl, etc.).
           For captures, use screenshotPath from Phase 1 instead of url.
           Returns tile summary with metadata and content hints (no tile images).
           Use tilesDir + start/end to fetch only the tiles you need.

Stop after Phase 1 if you only need the screenshot (capture mode) or comparison data.

4 tiling presets available:
  - "claude": 1092px tiles, ~1590 tokens/tile
  - "openai": 768px tiles, ~765 tokens/tile
  - "gemini3": 1536px tiles, ~1120 tokens/tile
  - "gemini": 768px tiles, ~258 tokens/tile

Supports: local files (filePath), remote images (sourceUrl), data URLs, base64, and web page capture (url — Chrome required).
Tiles saved as WebP (default) or PNG. Auto-downscales images over 10000px by default.

TOKEN COST NOTE: The get-tiles mode returns image tiles as inline base64, consuming significantly
more tokens than typical text-only MCP tools. Each tile costs ~258-1590 tokens depending
on preset. Use the Phase 2 summary and tile hints to fetch only non-blank, relevant tiles.

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

```json
"mcpServers": {
  "image-tiler-mcp-server": {
    "command": "npx",
    "args": ["-y","image-tiler-mcp-server"]
  }
}
```

## Documentation

## What keiver/image-tiler-mcp-server MCP server does

The keiver/image-tiler-mcp-server MCP server exposes one MCP tool, `tiler`, for preparing large visual inputs for language models. It accepts local files, remote image URLs, data URLs, base64 content, and web-page URLs. For a web page, it uses Chrome to produce a full-page screenshot before processing it.

The server is suited to visual QA, responsive-layout checks, diagrams, infographics, design mockups, and other images where a model's normal resizing could remove useful detail. It can also return only a captured screenshot when tiling is not needed.

## How it works

Use the tool in two calls. The first call provides only the image source, such as `filePath`, `sourceUrl`, or `url`. It returns an output directory, and for image inputs it presents a comparison of the available presets with tile counts and estimated token use. The user must choose the preset rather than having the workflow silently select one.

The second call repeats the source and supplies the selected preset plus the output directory. For a web capture, use the screenshot path returned by the first call. This phase returns tile metadata, a summary, and content hints rather than embedding every tile immediately. Request only the needed tile range from the tile directory; tiles are returned as inline base64 images and can consume substantial context.

The four presets are `claude` with 1092-pixel tiles and about 1590 tokens per tile, `openai` with 768-pixel tiles and about 765 tokens, `gemini3` with 1536-pixel tiles and about 1120 tokens, and `gemini` with 768-pixel tiles and about 258 tokens. Images larger than 10,000 pixels are downscaled by default. Output is WebP unless PNG is requested.

## Setup and configuration

The keiver/image-tiler-mcp-server MCP server requires Node.js 20 or newer. The package can be run directly with `npx -y image-tiler-mcp-server`, installed globally with npm, built from the repository, or run in Docker. It uses the standard stdio transport, so the MCP client configuration supplies the command and arguments.

Named setup examples are provided for Claude Code, Codex CLI, VS Code with Cline or Continue, Cursor, and Claude Desktop. A Docker deployment must use interactive stdin, mount directories that the server reads or writes, and can set `TILER_ALLOWED_DIRS` to limit accessible paths. Docker includes Chromium for URL capture. `TILER_DISABLE_URL_CAPTURE=1` disables that feature, while `CHROME_NO_SANDBOX=0` re-enables Chrome's sandbox when the container environment supports it.

## Tools and capabilities

- `tiler`: tile an image or capture and tile a web page.
- Compare model presets before committing to a tiling strategy.
- Classify tiles by visual detail, including blank, low-detail, mixed, and high-detail results.
- Generate an interactive HTML preview with model-specific grids and token estimates.
- Capture full pages by scrolling through lazy-loaded content and stitching sections beyond 16,384 pixels.
- Capture mobile layouts using a 390-pixel viewport, 2x retina scale, and a mobile Safari user agent.
- Save results in versioned output directories for repeated capture-and-review workflows.

## Limitations and notes

URL capture requires Chrome. The OpenAI preset targets the GPT-4o and o-series vision pipeline; GPT-4.1 is not supported by that configuration. Gemini 3 has a fixed token budget per image, so tiling can increase total token use even though it preserves local detail.

Do not request every tile by default. The second-phase summary and content hints are intended to help the agent skip blank or irrelevant regions. Each returned image tile can cost roughly 258 to 1590 tokens depending on the selected preset.

_Full upstream README: https://allmcps.com/mcp/keiver-image-tiler-mcp-server/readme_