cf_ai_canvas
An AI-powered collaborative canvas with remote MCP server on Cloudflare Workers.
Power your agent
การติดตั้งจากผู้เผยแพร่ ไม่ใช่เอนด์พอยต์ของ PlugMCP
URL แบบรีโมตหรือคำสั่งบนเครื่องของคุณ? การติดตั้งสองแบบต่างกันอย่างไร
รันบน PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
เปิดจากผู้เผยแพร่
An AI-powered collaborative canvas with remote MCP server on Cloudflare Workers.
AI-powered collaborative canvas and remote MCP server on Cloudflare Workers
| Endpoint |
|---|
| URL |
|---|
| App | https://cf-ai-canvas.mc146.workers.dev |
| MCP endpoint | https://cf-ai-canvas.mc146.workers.dev/mcp |
flowchart LR
user["User"]
mcpClient["External MCP client\nClaude / Cursor / VS Code"]
frontend["React + tldraw v5\nCloudflare Assets"]
worker["Cloudflare Worker\nsrc/server.ts"]
chat["ChatAgent DO\nAIChatAgent — chat + canvas state"]
mcp["CanvasMCP DO\nRemote MCP server at /mcp"]
aiText["Workers AI\nLlama 3.3 70B (text)"]
aiVision["Workers AI\nLlama 3.2 11B Vision"]
kv["Workers KV\nNamed snapshots"]
user -->|"prompt / image"| frontend
frontend <-->|"WebSocket state sync"| chat
frontend -->|"HTTP / WebSocket"| worker
worker --> chat
worker --> mcp
chat -->|"diagram planning"| aiText
chat -->|"image analysis"| aiVision
aiVision -->|"description"| aiText
mcpClient -->|"Streamable HTTP MCP"| mcp
mcp -->|"snapshot / restore"| kv
Click any of the three preset buttons to instantly send a diagram request:
Each button auto-sends on click. No need to press Send.
Type any diagram description in the chat box and press Send. Examples:
Draw a microservices architecture with API gateway, auth service, and database
Create a data pipeline with ingestion, validation, transformation, and warehouse
Draw a Kubernetes cluster with ingress, deployments, services, and persistent volumes
Click Attach Image, select a PNG/JPG (hand-drawn sketch, existing diagram, screenshot). The app uses Llama 3.2 Vision to extract structural context from the image, then uses that to generate a clean tldraw diagram. Works best with architecture sketches and flowcharts.
git clone https://github.com/Mihai-Codes/cf_ai_canvas.git
cd cf_ai_canvas
npm install
# One-time KV namespace setup
npx wrangler kv namespace create "CANVAS_KV"
# Update the KV ID in wrangler.jsonc
npx wrangler login
npm run dev
# App: http://localhost:8787
# MCP: http://localhost:8787/mcp
npm run deploy
# Deploys to https://cf-ai-canvas.mc146.workers.dev
npx @modelcontextprotocol/inspector@latest
# Transport: Streamable HTTP
# URL: https://cf-ai-canvas.mc146.workers.dev/mcp
cf_ai_canvas/
├── src/
│ ├── server.ts # Worker entry point + routing
│ ├── chat-agent.ts # AIChatAgent — NL/image → diagram
│ ├── canvas-mcp.ts # McpAgent — 17 canvas tools at /mcp
│ ├── client.tsx # React + tldraw v5 frontend
│ ├── diagram-patterns.ts # Deterministic pattern library for 5 diagram types
│ ├── types.ts # Shared TypeScript types
│ └── styles.css # App styles + tldraw arrow label overrides
├── test/
│ └── diagram-scenarios.spec.ts # Playwright e2e — all 3 input paths
├── test-results/ # Playwright screenshots
├── .github/workflows/ci.yml # CI: typecheck + build + deploy
├── PROMPTS.md # AI-assisted development log
└── wrangler.jsonc # Cloudflare bindings configuration
All 7 tests pass against https://cf-ai-canvas.mc146.workers.dev in ~50s:
npm test
GitHub Actions runs on every push to main:
tsc --noEmit)vite build)npx wrangler deploy — now blocking to ensure successful deployment, requires Zone:Read on the API token)Required secrets: CLOUDFLARE_ACCOUNT_ID, CLOUDFLARE_API_TOKEN
If the canvas appears empty or disappears after a page refresh:
caches storage on load, reducing stale asset issues.Version ID printed in the console).AI-powered collaborative canvas and remote MCP server on Cloudflare Workers
เครื่องมือ
ยังไม่มีรายการเครื่องมือ เซิร์ฟเวอร์แบบรีโมตจะได้รายการหลังการตรวจสอบสถานะครั้งแรก
FAQ
คัดลอกคอนฟิก stdio สำหรับ Cursor, Claude Desktop, Claude Code หรือ VS Code ได้ในหน้านี้ คอนฟิกนี้ใช้การติดตั้งของผู้เผยแพร่ ไม่ใช่พร็อกซีของ PlugMCP
ระเบียนนี้ไม่ได้ระบุตัวแปรสภาพแวดล้อมที่จำเป็น
Cf Ai Canvas ทำงานแบบโลคัลผ่าน stdio
ตรวจสอบล่าสุดเมื่อ 2026-10-28 AI-powered collaborative canvas and remote MCP server on Cloudflare Workers