Three MCP tools ยท stdio transport

Everything your model couldn't see

Three registered MCP tools give your AI real perception ... screenshots analyzed as structured JSON, source trees read safely, and ZIP archives inspected in memory. No special folders, no setup rituals.

analyze_images

One reference or an ordered array of images. Every reference type is resolved automatically ... absolute paths, relative paths, file:// URLs, Markdown image links, inline base64 data URLs, OpenCode caches, and bounded temporary-folder scans.

Images are processed sequentially in the exact supplied order. A failed image gets its own result and never discards the rest.

structured UI JSONordered batchper-image resultAI + OCR fallback
Example call ... analyze screenshots
{
  "image_paths": [
    "C:/Screenshots/home.png",
    "C:/Screenshots/settings.png"
  ],
  "prompt": "Focus on responsive layout,
  reusable components, and accessibility."
}

read_code_context

Point it at any file or folder and get a clean file tree followed by safe text contents. node_modules, .git, build output, binaries and common secret files are skipped automatically, and output is capped so the model never drowns.

Optional max_files and max_output_bytes keep every request predictable.

file tree + contentssecret-file filteringsize limits
Example call ... read a project folder
{
  "path": "C:/Projects/my-app/src",
  "max_files": 80,
  "max_output_bytes": 750000
}

read_zip_context

Inspect any .zip entirely in memory ... a virtual file tree plus readable contents, with zero extraction to disk. Path traversal, duplicates, encryption, CRC, compression-ratio and ZIP-bomb checks run on every archive.

Safe enough to hand your model an archive you barely trust.

in-memory onlyZIP-bomb safeno extraction
Example call ... read a ZIP safely
{
  "path": "C:/Downloads/my-app.zip",
  "max_files": 80,
  "max_output_bytes": 750000
}

Capabilities

Under the hood, everything counts

๐Ÿ–ผ๏ธ

Image context

Screenshots, UI mocks, diagrams ... analyzed in order, one result per image.

๐Ÿ”

Attachment recovery

Paths, URLs, markdown links, inline data & bounded temp-folder scans.

๐Ÿงฑ

UI analysis

Structured JSON: layout, components, typography, colors, spacing, interactions, a11y.

๐Ÿ“

Code context

Clean file tree + safe source contents, filtered and size-limited.

๐Ÿ—œ๏ธ

ZIP context

In-memory archive inspection ... zero extraction, full security checks.

๐Ÿ“

Local measurements

Resized dimensions, dominant color and palette measured with sharp.

๐Ÿง 

Automatic vision

OCR by default; add Gemini or OpenAI later ... cloud failures fall back to OCR.

๐Ÿงฉ

Universal

One MCP server for Codex, OpenCode, Claude Code, Cursor, VS Code & Claude Desktop.

Ready to see everything?

Setup takes under a minute and works with zero API keys.