OpenCode

View as Markdown

OpenCode is an open-source, terminal-based AI coding agent. It works with the full context of your project - reading across the codebase, running shell commands, editing files wherever changes are needed, and taking care of git along the way - and it's model-agnostic, so you can pair it with the provider of your choice. Paired with Buddy, OpenCode can deploy your project and let Buddy handle the entire deployment workflow. On top of that, it can trigger and monitor your pipelines, ship changes across environments, and dig through build output and logs when something needs fixing.

Info
This page covers working with Buddy from within OpenCode. On top of that, you can run AI coding agents in a fully isolated Buddy sandbox - see Run agents in a sandbox.

Quick start

1

Install the Buddy CLI

bash
npm install -g bdy
$
2

Log in to Buddy

bash
bdy login
$
3

Link your project

Run this inside your project directory to connect it to a Buddy project:

bash
bdy proj link
$
4

Install the Buddy skill

bash
npx skills add buddy/buddy-plugin
$

The Buddy skill teaches the agent to drive Buddy through the bdy CLI - link a project, deploy, run pipelines, ship across environments, and read build logs - so you can ask for all of it in plain language.

5

Start OpenCode

bash
opencode
$
6

Prompt the agent

Ask it to build and ship something end to end, for example:

prompt
Create a snake app and deploy it for me

Example prompts

Once the agent is running and your project is linked to Buddy, try prompts like:

  • Create a Next.js app and deploy it to production.
  • Deploy an Umami analytics instance and wire it into my app.
  • Deploy an AI chatbot with Postgres for history and Redis for rate limiting.
  • Deploy my app, set up a CDN, and buy and configure a domain with SSL.
  • Add a pipeline that runs on every change in the repository and executes my test suite.
  • Add a pipeline that runs every minute, checks that my site is up, and sends a notification to Slack, email, Telegram, or SMS if anything is wrong.
  • Check the last few CI pipeline runs for failures, read the logs, and fix the failing tests.
  • Analyze the logs from the last production deployment and fix the errors.
  • Find the last successful production deployment and roll back to it.
  • Check my app logs and open issues based on what you find.

Connect MCP

Give OpenCode access to Buddy over MCP so it can run pipelines, manage deployments, and read logs. See the MCP page for the regional server URLs, authentication modes, and the full list of tool categories.

With the remote server

Add the server to opencode.json:

json
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "buddy": {
      "type": "remote",
      "url": "https://mcp.buddy.works/mcp",
      "enabled": true
    }
  }
}

Or ask OpenCode to set it up for you:

prompt
Add the Buddy MCP server (https://mcp.buddy.works/mcp) to your MCP configuration.

With the Buddy CLI

If you're already logged in with bdy login, bdy mcp runs a local MCP server that proxies to Buddy using your CLI session - no OAuth prompt, no token to manage. Add it to opencode.json as a local server:

json
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "buddy": {
      "type": "local",
      "command": ["bdy", "mcp", "--tools", "pipelines,source"],
      "enabled": true
    }
  }
}

Omit the --tools arguments to expose all categories.

Last modified on Oct 1, 2026