<p align="center"> <img src=".github/og.png" alt="claude-image-tools — Image generation MCPs and Skills, in one place" width="100%"> </p>

claude-image-tools

English · 日本語

Image generation in Claude Code, all in one place — just ask. An operations hub that wires up image-generation MCPs and Skills for Claude Code on Windows, so OG banners, diagrams, posters, and GIFs become "just ask Claude" deliverables.

What you get in 30 seconds

What you can ask Claude Code, and which tool runs underneath:

Architecture

flowchart LR
    user([User]) --> cc[Claude Code]
    cc -->|"Text → image<br/>(strong CJK typography)"| openai[openai-image MCP<br/>gpt-image-2]
    cc -->|"Hand-drawn diagrams"| excal[excalidraw MCP]
    cc -->|"Mermaid diagrams"| mermaid[claude-mermaid MCP]
    cc -->|"Print-quality / art / GIF"| skills[Anthropic Skills<br/>canvas-design / algorithmic-art<br/>slack-gif-creator]
    openai --> out[(Pictures/claude-generated/)]
    excal --> out
    mermaid --> out
    skills --> out

How this differs from nearby choices

Quick start (Windows)

Prerequisites: Claude Code, Node.js, Python (uv), and OPENAI_API_KEY set as a Windows user environment variable.

# 1. openai-image MCP (gpt-image-2)
uv tool install openai-gen-image-mcp
claude mcp add --scope user openai-image `
  -e OPENAI_API_KEY=$env:OPENAI_API_KEY `
  -- C:/Users/<your-username>/.local/bin/openai-gen-image-mcp.exe

# 2. claude-mermaid (Mermaid diagrams)
npm install -g claude-mermaid
claude mcp add --scope user mermaid -- claude-mermaid

# 3. excalidraw MCP (hand-drawn diagrams)
git clone https://github.com/yctimlin/mcp_excalidraw
cd mcp_excalidraw
npm install; npm run build
claude mcp add --scope user excalidraw -- node "$PWD/dist/index.js"
# When using it, run the canvas server in a separate process: npm run canvas (port 3000)

# 4. Bulk-install Anthropic Skills
claude plugin install example-skills@anthropic-agent-skills

Replace <your-username> with your Windows account name (the value of %USERNAME%).

Verify connection:

claude mcp list
# Success when excalidraw / openai-image / mermaid all show ✓ Connected

Smoke-test prompts

Try these in a fresh Claude Code session. Replace <your-username> with your Windows account name, or point output_path at any folder you can write to (e.g. ~/Pictures/claude-generated/):

Use the openai-image MCP to generate a 1024x1024 image.
Prompt: "a small cute cat sitting on a wooden table, soft lighting".
output_path: C:/Users/<your-username>/Pictures/claude-generated/test-cat.png

Generate a 1200x630 image. Center a large "Caveat" title,
with a small Japanese subtitle "罠を記録する OSS" beneath it,
on a deep blue gradient background.
output_path: C:/Users/<your-username>/Pictures/claude-generated/test-og.png

If the text — Japanese in particular — renders cleanly, you're set.

Windows-specific gotchas (you will hit these)

This stack hand-patches a few upstream packages so they run on Windows. See the checklist in CLAUDE.md for exact patch points. Common symptoms:

• mermaid_preview errors with spawn npx ENOENT → claude-mermaid's spawn calls are rewritten to go via cmd.exe /c
• openai-image dies right after launch → pass the API key through Claude Code config (Windows env-var propagation timing issue)
• excalidraw's export_to_image rejects paths → it only writes under CWD; change the allowed base via EXCALIDRAW_EXPORT_DIR

<details> <summary>Common OpenAI billing snags</summary>

• ChatGPT Plus and pay-as-you-go are separate wallets: a Plus subscription does not unlock the image API
• gpt-image-2 requires organization verification: unverified orgs get a 403; verification takes up to 15 minutes
• Zero balance returns billing_hard_limit_reached (HTTP 400) — check the billing console first

</details>

<details> <summary>Design choices for this repo</summary>

• No application code lives here: it's purely an operations hub for the MCP / Skill stack
• Push fixes upstream: root-cause fixes for the gotchas go to each upstream as PRs/Issues (notes accumulate via the caveat MCP)
• Dogfooding: the OG banner of this repo was generated by openai-image itself

</details>

Output location

Default save path for generated images: C:\Users\<your-username>\Pictures\claude-generated\. New tools should follow the same convention.

Contributing

Issues and PRs are welcome. Since the gotchas this hub documents are upstream bugs, upstream fixes are preferred — open a PR/Issue against the affected package (claude-mermaid, mcp_excalidraw, openai-gen-image-mcp) and link it here so the workaround can be retired.

License

MIT