Build with your coding agent

Rubry's docs are also an MCP server, at https://docs.rubry.ai/mcp. Add it to your coding agent (Claude Code, Cursor, VS Code, Codex, Gemini CLI, Windsurf, the Claude app…) and ask for what you want:

Build me a support agent that answers from our FAQ page and looks up orders in our API, deploy it on my Cloudflare account and connect it to rubry.ai.

Your coding agent reads these docs, writes the agent and its tools in the template, and follows the deploy steps. Add Cloudflare's MCP server too, and it also sets up the Cloudflare side (databases, secrets, domain, logs) without you opening the dashboard or learning Wrangler.

It is free, read-only and needs no key: it only returns docs and code. Everything it suggests runs on your machine and your accounts, with your approval.

1. Add the servers

Add Rubry (https://docs.rubry.ai/mcp) and, for the infrastructure, Cloudflare (https://mcp.cloudflare.com/mcp, which asks you to sign in to Cloudflare the first time).

Claude Code

claude mcp add --transport http rubry https://docs.rubry.ai/mcp
claude mcp add --transport http cloudflare https://mcp.cloudflare.com/mcp

Cursor

In ~/.cursor/mcp.json (or .cursor/mcp.json in the project):

{
  "mcpServers": {
    "rubry": { "url": "https://docs.rubry.ai/mcp" },
    "cloudflare": { "url": "https://mcp.cloudflare.com/mcp" }
  }
}

VS Code (Copilot)

In .vscode/mcp.json:

{
  "servers": {
    "rubry": { "type": "http", "url": "https://docs.rubry.ai/mcp" },
    "cloudflare": { "type": "http", "url": "https://mcp.cloudflare.com/mcp" }
  }
}

Codex

In ~/.codex/config.toml:

[mcp_servers.rubry]
url = "https://docs.rubry.ai/mcp"

[mcp_servers.cloudflare]
url = "https://mcp.cloudflare.com/mcp"

Gemini CLI

In ~/.gemini/settings.json:

{
  "mcpServers": {
    "rubry": { "httpUrl": "https://docs.rubry.ai/mcp" },
    "cloudflare": { "httpUrl": "https://mcp.cloudflare.com/mcp" }
  }
}

Windsurf

In ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "rubry": { "serverUrl": "https://docs.rubry.ai/mcp" },
    "cloudflare": { "serverUrl": "https://mcp.cloudflare.com/mcp" }
  }
}

Claude (web and desktop)

Settings → Connectors → Add custom connector, with the URL https://docs.rubry.ai/mcp. Do the same with https://mcp.cloudflare.com/mcp.

Any other client that supports remote MCP servers over HTTP ("Streamable HTTP") works the same way.

2. Ask for your agent

Describe the agent in plain words: what it is for, what it can look up or do, and where it should run. For example:

• "Create a Rubry agent that writes Instagram captions in our brand voice, and run it locally."
• "Add a tool to my researcher agent that reads our Notion page, with the token as a secret."
• "Deploy my rubry template on Cloudflare with the domain agents.example.com, then give me the endpoint and a key to paste on rubry.ai."

Clients that support MCP prompts also offer create-agent: pick it, describe the idea, and it walks the whole way from template to rubry.ai.

What the Rubry server offers

Tool	What it does
search_docs	Searches these docs, in English or Portuguese
read_doc	Returns a page in Markdown (quick start, agents and tools, deploy…)
list_docs	Lists the pages
scaffold_agent	Writes the files for a new agent and its tools, following the template's conventions, plus the line to add in agents/src/index.ts and the secrets to set
deploy_plan	The deploy steps for your case (with or without Cloudflare's MCP, custom domain, several assistants), in order

The pages are also available as MCP resources (rubry://docs/<page>).

What Cloudflare's MCP server does

Cloudflare's MCP server gives your coding agent the whole Cloudflare API and Cloudflare's docs, under your account and the permissions you grant when you sign in. With it, the agent can:

• find your account and zones;
• create the two D1 databases the template needs and fill in their ids;
• attach a custom domain to the middleware;
• read the Workers' logs when something fails.

Uploading the code itself still goes through npm run deploy (Wrangler), because the Workers need to be bundled first; your agent runs it for you.

In CI, or without a browser, pass a Cloudflare API token as a bearer token instead of signing in.

:::tip Keep secrets out of the chat Secrets your agent creates, like SECRET_KEY, are best generated and set in one command, so they never pass through the conversation: openssl rand -hex 32 | npx wrangler secret put SECRET_KEY. The deploy plan does it that way. The ADMIN_TOKEN is the one you need to know: keep it in your password manager. :::

Going deeper with Mastra

The template's agents are Mastra agents, so anything Mastra does is open to you. For your coding agent to know Mastra's full toolkit (workflows, memory, evals, MCP clients…), give it Mastra's docs too:

• Mastra's skill, Mastra's recommended way: see Develop with AI.
• Mastra's MCP docs server, run locally: npx -y @mastra/mcp-docs-server@latest (in Claude Code: claude mcp add mastra -- npx -y @mastra/mcp-docs-server@latest). If you can't run a local server, there is a remote one at https://mastra.mcp.kapa.ai (sign-in with Google, for rate limits).
• Plain text: mastra.ai/llms.txt lists every page as Markdown.

Your agents can also use other MCP servers as tools: Mastra's MCPClient (@mastra/mcp) connects an agent to remote MCP servers over HTTP, such as a CRM's or a database's. On Workers, use remote (HTTP) servers: local stdio servers can't run there.

Prefer not to do it yourself?

Rubrion builds and runs agents for companies, on this same stack.