Resources with an identity
Lambda is a function. DynamoDB is a database. Provider catalogs give each node more than a label.
Built for LLMs and MCP
Ask Claude, Cursor or Codex to map AWS, Cloudflare, GCP and Kubernetes. Get checked source, a rendered diagram and a link to keep editing in the playground.
Free to use. No API keys or account required.
Diagram Cloudflare Workers proxying requests to AWS Lambda and publishing to GCP Pub/Sub, with both paths connected to a Kubernetes workload. Validate it and give me a playground link.
Describe your system. The agent skill supplies the language, provider rules and examples.
Cloud catalogs, validation and rendering, with tools, prompts and documentation ready for your agent.
Open the same source in the playground. Inspect, edit and export your architecture.
Connect the free MCP server or install the skill. Let your LLM draw, then inspect and refine the same source in the playground.
Give your agent cloud catalogs, semantic validation, diagram rendering and editable playground links. Works with any MCP-compatible client.
Client setup instructionshttps://mcp.archlex.dev/mcpFree to use. No API keys or account required.
Set up the ArchLex MCP server in the MCP-capable client or coding-agent environment I am currently using (Claude Code, Cursor, Codex, VS Code/Copilot, or another compatible client). Use the remote Streamable HTTP endpoint https://mcp.archlex.dev/mcp and name the server archlex. It is free to use and requires no API key, account, or authentication headers. If this environment does not support MCP, explain that and suggest a compatible client. Otherwise, detect this client's supported setup method and configuration scope. Add or update only the ArchLex entry; preserve all other MCP servers and settings. If you cannot access the configuration, give me the exact client-specific steps instead of claiming it is configured. Connect to the server and verify that render_diagram, validate_diagram, get_cloud_catalog, and generate_playground_url are available. If a client restart or permission approval is needed, explain what I need to do and verify again afterward. Once connected, render a small example using render_diagram, show the result if supported, and provide the returned editable playground link. Report any setup or connection errors accurately.
Install the skill, then ask your agent to draw your architecture. It gets the DSL reference, provider rules and working examples.
Agent setup guidenpx skills add baires/archlexOpen the same multicloud source in the playground. Inspect diagnostics, edit relationships and export SVG. Start directly in your browser, with no installation.
Open multicloud playgrounddirection LR
edge: cloudflare.workers["Cloudflare Workers"]
api: aws.lambda["AWS Lambda"]
events: gcp.pubsub["GCP Pub/Sub"]
cluster production {
namespace orders {
app: k8s.deployment["Kubernetes workload"]
}
}
edge -[proxies]-> api
edge -[publishes]-> events
api -[connects]-> app
events -[connects]-> appUse the same engine in Node.js or the browser. Compose the providers you need and keep the source alongside your code.
Explore the APInpm install @archlex/core @archlex/aws @archlex/cloudflare @archlex/gcp @archlex/k8sAn LLM workflow, with a human workspace. Explore the playground
Resources have an identity. Connections have rules. Give your agent a language that understands both.
Try this AWS example: change a supported connection, inspect the real provider warning, then correct it.
Explore the agent workflowprovider aws
direction LR
api: api-gateway["API Gateway"]
worker: lambda["Order handler"]
data: dynamodb["Orders"]
api -[invokes]-> worker
worker -[writes]-> dataAWS-RELATIONSHIP-INVALID-ENDPOINT-001Relationship 'invokes' is not supported to 'dynamodb' (expected: lambda, sns, sqs, step-functions, step-functions-express, api-gateway, ecs, eks).
Use a supported target for 'invokes' or choose a different relationship kind.
A line is easy to draw. ArchLex checks what it connects, where resources belong, and what the relationship says.
Lambda is a function. DynamoDB is a database. Provider catalogs give each node more than a label.
“Invokes” and “writes” mean different things. Supported endpoints are checked against provider declarations.
Accounts, regions, VPCs, subnets and namespaces describe structure. ArchLex checks supported placement rules.
api: api-gateway
worker: lambda
data: dynamodb
api -[invokes]-> worker
− worker -[invokes]-> data
+ worker -[writes]-> dataYour diagram lives in plain text. Review the relationship that changed, keep the source in Git, and render it as SVG.
Make your first changeProvider-aware resources and official icons.
Mix providers in one diagram when your architecture does.
Accounts, regions, VPCs and subnets.
194 servicesReferenceEdge services, accounts and origin connections.
92 servicesReferenceGoogle Cloud resources and typed relationships.
189 servicesReferenceClusters, namespaces and workload relationships.
62 resourcesReferenceDeclared diagram rules: known resources, supported scopes and typed relationship endpoints. It checks your architecture description against provider catalogs. It does not inspect a deployed system or certify its security or availability.
Yes. ArchLex is built for LLM workflows: connect the free hosted MCP server or install the agent skill. Claude Code, Cursor, Codex and other compatible clients can look up resources, validate source, render diagrams and return an editable playground link.
The hosted MCP server is free to use, with no API keys or account required. The playground is free too. ArchLex is open source under the MIT license, and you can also run the CLI or packages locally.
Yes. Write a diagram in the playground, render it with the CLI, or embed @archlex/core in a browser or Node.js application. The source is plain text and the output is SVG.
Bring your agent. The MCP server and playground are ready.