OpenCode Crash Course Automate Workflow with Free AI Models

Build autonomous coding agents with OpenCode - MCP, Skills, Sub-agents, Plugins & AGENTS md. No paid AI subscription.
What you'll learn

---

Run OpenCode entirely free using Zen models, or configure a paid provider in three minutes - provider-agnostic across 75+ models
Master built-in slash commands and build your own custom commands for repeatable, shareable workflows
Write and test a project-aware AGENTS md file so OpenCode understands your codebase, conventions, and constraints
Understand MCP (Model Context Protocol) from first principles - and connect OpenCode to GitHub, databases, and Context7
Configure OpenCode's allow, ask, and deny permission model to control exactly what your agents can touch
Build production Skills with YAML front-matter and Markdown - including a Data Cleaning Skill and a Dashboard Creation Skill
Design custom sub-agents including a working Code Reviewer, and know exactly when to use a Skill vs a sub-agent
Requirements

---

No paid AI subscription required - every lesson runs on free models
No prior experience with AI coding agents, OpenCode, or MCP is required - we start from the fundamentals
A computer (Windows, macOS, or Linux) with internet access
Description

---

Stop paying for AI coding subscriptions. Start running autonomous agents for free.
OpenCode is the open-source AI coding agent built for the terminal - provider-agnostic, MIT-licensed, and able to run entirely on free AI models. This crash course takes you from installation to a packaged, shareable agent toolkit in under four hours.
This is not a course about typing prompts and hoping for the best. You will build a complete, reproducible automation system: custom slash commands, a project-aware AGENTS md file, MCP connections to GitHub and your database, two production Skills, a custom sub-agent, and your own distributable OpenCode plugin.
What you will build
Every hands-on lesson runs against one real case study - a messy retail sales dataset with duplicate orders, inconsistent date formats, currency-formatted text, and missing regions. You will clean it, analyse it, and dashboard it using agents you build yourself.
AData Cleaning Skill that handles real-world data defects through rules you write
ADashboard Creation Skill that turns clean data into visual reporting
ACode Reviewer sub-agent that audits your work automatically
Ashareable plugin bundling your skill, agent, and command into one installable package
What You Will Learn

---

How the OpenCode agentic loop works internally: read, reason, act, verify
Free setup using Zen models, plus paid configuration in three minutes if you want it
Plan mode vs Build mode, and why separating reasoning from execution reduces errors
Context windows, auto-compaction, and how to keep agents on track in long sessions
Writing and testing an AGENTS md file so your agent understands your project
MCP from first principles: GitHub, database servers, Context7, and the allow/ask/deny trust model
Skills architecture: YAML front-matter and Markdown body, plus installing from the marketplace
Sub-agents vs Skills, and exactly when to reach for each
Packaging and distributing your work as an OpenCode plugin
Who this course is for

---

Developers who want agentic coding in the terminal without paying for a subscription - OpenCode runs free on open models
Claude Code, Cursor, or Copilot users who want a provider-agnostic, open-source alternative they fully control
Data analysts and engineers who want to automate cleaning, analysis, and reporting workflows with agents they build themselves
Homepage

---