Guides for agents
Task guides for AI agents and the people who run them: run code in a cloud sandbox, search the web, deploy a website, schedule an agent, call any model, generate media, and use API keys safely.Each guide answers one task an agent is asked to do, with the steps for the Dexto CLI, the Dexto MCP server and the HTTP API.
| Task | Guide |
|---|---|
| Run code, tests or a build somewhere other than your own machine | Run code in a cloud sandbox |
| Look something up on the web and read the pages | Search the web and read pages |
| Put a page or a small site online at a public URL | Deploy a website |
| Do a job every morning, at a set time, or when a webhook fires | Run an agent on a schedule |
| Use Claude, GPT, Gemini, Grok and others with one key | Call any model with one API |
| Make an image, a video or speech | Generate images, video and audio |
| Let an agent call an API with a key it never sees | Use an API key without exposing it |
Before you start
Section titled “Before you start”Every guide assumes the CLI is installed and signed in:
npm install -g dextodexto login --devicedexto whoami --jsondexto login --device prints a link and a code; a person approves it. dexto whoami --json
exits 0 once you are signed in, and exits 3 with needs_sign_in until then. On a server or in CI,
set DEXTO_API_KEY to a developer key of your Cloud project instead. See
Dexto CLI for agents for exit codes and the rules an agent follows.
For MCP, add https://app.dexto.ai/mcp to your client and sign in once; see
Connect via MCP. Not every step has an MCP tool: some stay with the CLI and the API on
purpose, and each guide says which.
Paid steps (sandboxes, search, models and media) are charged to your Cloud project’s credits.