figurify-mcp 0.2.1

Ship-ready MCP server for Gemini + Imagen diagram/plot generation. Runs on Claude, ChatGPT, VSCode Copilot, Cursor.

figurify-mcp

MCP server for Google Gemini + Imagen image generation. Works with Claude Desktop, Claude Code, Cursor, VSCode Copilot, and any MCP-compatible host.

• 5 generation tools — diagrams, infographics, paper figures, matplotlib plots, VLM judging
• Smart model routing across Gemini 2.5 / 3 / 3.1 + Imagen 4 / Nano Banana
• 1K by default to stay under Claude's 1 MB inline tool-result cap
• Stdio + streamable HTTP transports

---

Install

pip install figurify-mcp        # or: uv pip install figurify-mcp

Get a free Gemini API key: aistudio.google.com/apikey

Configure your host

Claude Desktop — ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "figurify": {
      "command": "uvx",
      "args": ["figurify-mcp"],
      "env": { "GOOGLE_API_KEY": "your-key" }
    }
  }
}

Claude Code / Cursor

claude mcp add figurify -s user \
  -e GOOGLE_API_KEY=$GOOGLE_API_KEY \
  -- uvx figurify-mcp

VSCode Copilot — .vscode/mcp.json

{
  "servers": {
    "figurify": {
      "command": "uvx",
      "args": ["figurify-mcp"],
      "env": { "GOOGLE_API_KEY": "your-key" }
    }
  }
}

Restart the host. The tools appear automatically.

---

Tools

Tool	What it does	Typical cost
create_simple_diagram	Quick prompt → image	~$0.04
create_infographic	Topic → styled poster (VLM layout + Imagen render)	~$0.12
create_methodology_diagram	Paper figure with N critique rounds	~$0.25
create_statistical_plot	Data → matplotlib chart via VLM codegen	~$0.002
evaluate_diagram	VLM-as-judge comparison of two diagrams	~$0.015

Meta (save host context): search_tools, describe_tool, recommend_model, estimate_cost, list_available_models.

Prompts: draft_methodology_figure, plan_visual_output.

Resources: figurify://styles/methodology_guidelines, figurify://styles/infographic_guidelines, figurify://catalog/models.

Images return as MCP ImageContent blocks with audience: ["user"] — they render inline in every compliant host.

---

Configuration

All env vars are optional.

Variable	Default	Purpose
GOOGLE_API_KEY	(required)	Gemini + Imagen API key
FIGURIFY_OUTPUT_DIR	~/.figurify/outputs	Where generated PNGs are saved
FIGURIFY_VLM_MODEL	(auto-routed)	Pin a specific VLM model
FIGURIFY_IMAGE_MODEL	(auto-routed)	Pin a specific image model
FIGURIFY_MAX_METHODOLOGY_ITERATIONS	4	Upper bound for critique loops

Transports

figurify-mcp                                            # stdio (default)
figurify-mcp --transport http --host 0.0.0.0 --port 8765

---

Security

This server runs VLM-generated matplotlib code in a subprocess. Shipped mitigations:

• AST + pattern allowlist rejects any code importing os, subprocess, socket, urllib, requests, pickle, etc.
• python -I isolated mode, minimal env, POSIX setrlimit (512 MB RAM / 120 s CPU / 100 MB fsize)
• load_image restricts filesystem reads to {output_dir, /tmp, CWD} — no path traversal
• API-key patterns and home paths stripped from every error message

Report vulnerabilities privately: [e-mail]. Full threat model in SECURITY.md.

Development

git clone https://github.com/genaimanoj/figurify-mcp
cd figurify-mcp
uv venv && source .venv/bin/activate
uv pip install -e '.[dev]'
cp .env.example .env   # set GOOGLE_API_KEY
pytest -q              # 64 tests, ~6 s

Contributions welcome — see CONTRIBUTING.md.

License

MIT © 2026 Manoj Bhandari. See LICENSE.