CloudSketch AI
Generate accurate C4, cloud, ERD, UML, sequence and process diagrams (Mermaid) from plain English.
Installation
Ajoutez ceci à la configuration MCP de votre client (Claude Desktop : claude_desktop_config.json ; Cursor : .cursor/mcp.json ; VS Code et les autres utilisent la même forme « mcpServers »).
cloudsketch-mcpclaude mcp add cloudsketch-ai -e 'CLOUDSKETCH_API_TOKEN=<your cloudsketch_api_token>' -- npx -y cloudsketch-mcp@0.1.0
{
"mcpServers": {
"cloudsketch-ai": {
"command": "npx",
"args": [
"-y",
"cloudsketch-mcp@0.1.0"
],
"env": {
"CLOUDSKETCH_API_TOKEN": "<your cloudsketch_api_token>"
}
}
}
}
Variables d'environnement
CLOUDSKETCH_API_TOKENsecret Personal API token from cloudsketchai.com (Account → API tokens). Needed for generate_diagram and check_credits.
Documentation
Tirée du README du projet sur GitHub, actualisée chaque semaine. Les liens ouvrent GitHub.
Sommaire
CloudSketch AI MCP server
Generate accurate architecture, database and process diagrams from Claude Desktop, Claude Code, Cursor, VS Code or any other MCP client, with CloudSketch AI.
Ask your assistant things like:
"Draw a C4 container diagram of our order service: React app, Node API, Postgres, Redis cache and Stripe." "Make an ERD for a clinic booking system with doctors, patients, slots and payments."
You get Mermaid code back, plus a link that opens the diagram in the CloudSketch AI editor. From there you can restyle it or export it to PowerPoint, Word, PNG, SVG, PDF or draw.io.
How accurate is it? Every diagram type is tested on fresh prompts, and the method and results are public: cloudsketchai.com/benchmark.
Tools
| Tool | What it does | Credits |
|---|---|---|
generate_diagram | Takes a description, a diagram type and an audience, and returns Mermaid code plus an edit link. | 1 (2 for C4, mind maps and technical sequence diagrams) |
open_in_cloudsketch | Turns any Mermaid code into a link that opens it in the editor. The diagram travels inside the link and is not uploaded. | none |
check_credits | Shows your plan and the credits left today and this month. | none |
list_diagram_types | Lists the 18 types and what each is for. | none |
generate_diagram supports 18 diagram types: C4 (context, container, component), cloud deployment, network, ERD, class, sequence, state, data flow (DFD), flowchart, swimlane, use case, user journey, quadrant, mind map, Sankey, data chart, gitgraph and Gantt.
The audience can be technical (the default), managers (plain words) or investors.
Diagrams made through the server are saved to your CloudSketch AI workspace like any other.
1. Get an API token
- Sign in at cloudsketchai.com. A free account works: 20 credits to start, then 10 a month.
- Open Account → API tokens and create a token. It's shown once, so copy it.
A token can only generate diagrams with your credits. It can't open your saved diagrams, billing or settings. You can delete it there at any time.
2. Add the server to your client
Node.js 18 or newer is all you need; the server has no dependencies.
Claude Desktop (claude_desktop_config.json), Cursor (~/.cursor/mcp.json), Windsurf and most other clients:
{
"mcpServers": {
"cloudsketch-ai": {
"command": "npx",
"args": ["-y", "cloudsketch-mcp"],
"env": { "CLOUDSKETCH_API_TOKEN": "csk_your_token" }
}
}
}
Claude Code:
claude mcp add cloudsketch-ai -e CLOUDSKETCH_API_TOKEN=csk_your_token -- npx -y cloudsketch-mcp
VS Code (.vscode/mcp.json):
{
"servers": {
"cloudsketch-ai": {
"type": "stdio",
"command": "npx",
"args": ["-y", "cloudsketch-mcp"],
"env": { "CLOUDSKETCH_API_TOKEN": "csk_your_token" }
}
}
}
Without npm: clone this repository, then use "command": "node" and "args": ["/path/to/cloudsketch-mcp/server.mjs"].
Without a token, open_in_cloudsketch and list_diagram_types still work.
Settings
| Variable | Default | Meaning |
|---|---|---|
CLOUDSKETCH_API_TOKEN | (none) | Your personal API token |
CLOUDSKETCH_BASE_URL | https://cloudsketchai.com | Only for a local or test build |
Privacy
- The server sends your description to cloudsketchai.com to generate the diagram, the same as typing it on the website.
open_in_cloudsketchsends nothing: it only builds a link.- See the privacy policy.
Develop
npm test runs the protocol and tool tests against a fake API, with no network and no credits used.
The server speaks MCP over stdio (newline-delimited JSON-RPC 2.0). It supports protocol versions 2025-06-18, 2025-03-26 and 2024-11-05.
Support
support@cloudsketchai.com · cloudsketchai.com
MIT licensed.
Questions fréquentes
Qu'est-ce que CloudSketch AI ?
Generate accurate C4, cloud, ERD, UML, sequence and process diagrams (Mermaid) from plain English.
Comment installer CloudSketch AI ?
Il s'exécute sur votre machine. Dans Claude Code : claude mcp add cloudsketch-ai -e 'CLOUDSKETCH_API_TOKEN=<your cloudsketch_api_token>' -- npx -y cloudsketch-mcp@0.1.0. Pour les autres clients, collez la configuration JSON de cette page dans leurs réglages MCP.
Faut-il une clé API pour CloudSketch AI ?
Oui. Il attend CLOUDSKETCH_API_TOKEN, à renseigner dans sa configuration.
CloudSketch AI est-il open source ?
Oui. Son code est sur GitHub sous licence MIT.
Qui maintient CloudSketch AI ?
Il est publié par cloudsketchai. AI.tn le référence sans le maintenir ni le recommander.
Issu du registre MCP officiel. Être listé n'est pas une recommandation : vérifiez le code et les permissions d'un serveur avant de lui donner accès à vos données.