Cloud Code + MCP + Skills = Desarrollo Automático

High-level summary

The video is a practical walkthrough of using “Cloud/Cloud Code” (Anthropic-style developer assistant) and its three main extensibility mechanisms: MCPs (Model Context Protocol connectors), Skills (agent/skill packs), and Plugins. The presenter demonstrates setup, common commands, integrations, testing automation, UI-improvement workflows, and security best practices.

---

Key concepts & components

cloud.md context file

• Running /init analyzes a project and generates cloud.md — a persistent context file Cloud uses to understand the repo (commands, stacks, architecture, env vars, file layout, conventions).
• You can edit cloud.md to add project conventions (naming, build rules, response verbosity). Keep it updated so Cloud uses correct context.
• There is a cross-tool alternative ens.md (shared by multiple AI tools); Cloud does not (yet) read it by default.

MCPs (Model Context Protocol)

• Purpose: connectors that let Cloud access external services (Notion, Gmail, GitHub, calendar, databases, cloud providers, browser automation, etc.).
• Common commands demonstrated:
  • cloud mp add <url/command> — install an MCP
  • cloud mp-list — list installed MCPs
  • cloud mp remove <name> — remove an MCP
  • cloud mp-hel — list MCP-specific commands
• Authentication: many MCPs use OAuth/consent flows (example: Notion).
• Loading: recent Cloud versions load MCPs on-demand (multiple MCPs can be installed and selected when needed).
• Security risks:
  • Unofficial MCPs (community GitHub servers) can be risky — potential for data exfiltration, credential leaks, or PR-injection (hidden prompts causing malicious actions).
  • Recommendation: prefer MCPs published/verified by the Cloud vendor (Anthropic) or other trusted authors; review the repo/author before installing.

Security best practice: vet MCPs and prefer official/verified ones. Install project-level when untrusted.

Browser automation / E2E testing

• Chrome DevTools / Chrome-from-Tools MCP (or Playwright/other tools) can open a separate Chromium instance under AI control.
• Use cases: register users, run UI flows, capture snapshots/screenshots, inspect console/DOM, run DB migrations, iterate until fixes pass.
• Demonstrated loop: Cloud runs tests, finds failures (e.g., missing tables), runs migrations, retries, and produces a test report with steps and errors.

Skills (agent skills)

• Skills are Markdown-configured agents/guides that tell Cloud how to perform specialized tasks (UI design, security audits, SEO, QA, docs, etc.).
• They encode best practices, styles, checklists, and can be written/curated by domain experts.
• Installation and management:
  • Install per-project (recommended) or globally; prefer project-level install for untrusted skills.
  • /skills lists installed skills.
  • Many community skills exist (AON Cloud Skills repo & others). Verify authors and repo quality.
• Example: an Interface Design skill can improve a dashboard UI (colors, layout, skeleton loading, sample data, consistent styling) and references UX/UI principles and repos.

Plugins

• Plugin bundles can package multiple skills + MCPs (e.g., Frontend Design, “superpers” bundle).
• Installed via /plugin; options to install locally for the repo, for contributors, or system-wide.
• Plugins simplify onboarding multiple capabilities at once.

---

Useful commands & workflows

• Start Cloud with the dangerous-skip-permissions flag to avoid repetitive permission prompts during automated sessions.
• slash init or /init — create cloud.md context for the project.
• /mp and the cloud mp-* family — add/list/remove MCPs.
• /skills — list installed skills; use slash + skill name in chat to invoke a skill.
• /plugin — browse and install plugin bundles.
• /context (and session reset) — inspect or clear session context; export sessions to clipboard/file.
• Remote control: run the “remote control” command to accept remote instructions and have Cloud send commands to your local machine terminal — useful to run tasks remotely (for example, from a phone) while actions execute on your dev machine.
• Bash/terminal integration: prefix commands with ! (or use bash mode) to run shell commands (e.g., pwd, ls -a). Dragging file paths or using VS Code “copy path” helps point Cloud to files.
• Path notation inside Cloud:
  • @ denotes project root (for file imports/navigation).
  • . or .. for relative path movement.
• Resuming sessions: when Cloud suggests a reconnect command after exit, paste it to resume conversation history.

---

Best practices & recommendations

• Keep cloud.md up-to-date and add conventions to reduce repetitive prompts.
• Install MCPs and Skills at the project level unless you fully trust them system-wide.
• Prefer official/verified MCPs; review community repos before installing.
• Use separate browser instances for automated tests to avoid using personal session cookies.
• Use Skills to get consistent, human-reviewed guidance (UI, security, SEO).
• Use snapshots and test reports for reproducibility and documentation.
• Always validate third-party MCPs/Skills before installing.

---

Example integrations & repositories mentioned

• Notion MCP (demo: authenticate & create items/pages/tasks).
• Chrome DevTools / Chrome-from-Tools MCP (Google project) for browser automation.
• Playwright, Puppeteer-like projects (Playwright mentioned as supported).
• AON MCP Servers (repo listing MCPs/services).
• AON Cloud Skills (skill repository).
• Community skill marketplace (not filtered — vet content).
• Plugins such as Frontend Design, Superpers (bundles of skills + MCPs).

---

Product / hosting sponsor

• Cynote (platform for deploying full-stack apps):
  • Features: GitHub/GitLab integration, public URLs, custom domains, automatic HTTPS, WebSockets & HTTP/2, real-time logs, environment variables, 24/7 support.
  • Pricing examples quoted: DBs from ~$1/month, backend servers from $3/month, 7-day free trial.

---

Limitations & advanced topics (not fully covered)

• Agents/agent teams and higher-level frameworks on top of Cloud.
• Using external models in Cloud, CLI-based integrations, further CI/CD or deployment automation.
• Upcoming deeper coverage promised (PR injection details, advanced MCP creation, extended Cloud Code course).

---

Practical tutorial checklist (reproducible steps)

1. Run slash init in your project to generate cloud.md.
2. Edit cloud.md to add naming conventions, build rules, and response preferences.
3. Install an MCP: copy install command from the MCP repo and run cloud mp add ..., then authenticate via browser for services like Notion.
4. Verify with cloud mp-list and cloud mp-hel.
5. Install a Skill (project-level): copy install command from the skill repo, choose Cloud as target, install locally.
6. Invoke a skill by typing /skill-name or using slash commands in Cloud; target specific files using @ for root paths or by dragging file paths from VS Code.
7. For browser-based tests, install the Chrome DevTools MCP and run the test instruction; review snapshots and logs.
8. Use remote-control to send terminal commands remotely (authorize and confirm).
9. Always validate third-party MCPs/skills before installing.

---

Main speakers / sources (as identified in the video)

• Video narrator / channel author (unnamed; references links like fast.df/discord; offers Cloud Code course).
• Anthropic / Cloud (referred to as “Cloud” / Cloud Code — the provider of the assistant, MCP marketplace, and official MCPs).
• Notion (MCP example and demo).
• Chrome-from-Tools / Google (browser automation MCP).
• Community repositories: AON MCP Servers, AON Cloud Skills, GitHub community skill repos.
• Sponsor: Cynote (deployment platform referenced in the video description).