How do I use OpenCode and Claude Code together?

Keep the project's rules in one AGENTS.md, which both agents read: OpenCode reads it first, and Claude Code reads it when there is no CLAUDE.md, or through a CLAUDE.md containing @AGENTS.md. Give each agent its own git worktree and branch, register the MCP servers you need in both, and move work between them with a commit and a handoff note. OpenCode can also pick up Claude Code's own files, CLAUDE.md and its skills folders, which makes it an easy second agent for a Claude Code project.

People pair them for a simple reason. Claude Code runs on Claude, and OpenCode runs on whichever provider and model you configure, so the second agent brings a different model's judgment to the same code: a review, a second attempt at a stubborn bug, or a cheaper model for routine work.

Which files does each one read?#

Two details catch people out:

• The first file found wins in OpenCode. Adding an AGENTS.md to a project means OpenCode stops reading its CLAUDE.md. Put everything in AGENTS.md, and keep CLAUDE.md to the single line @AGENTS.md if you want one at all.
• OpenCode v2 reads only AGENTS.md. The CLAUDE.md fallback is a v1 behaviour, so a project that relies on it loses its rules after an upgrade. One AGENTS.md works in both. To turn the Claude Code compatibility off in v1, set OPENCODE_DISABLE_CLAUDE_CODE=1.

How do I give both the same tools?#

Each agent keeps its own MCP list, so add a server to both. In Claude Code:

claude mcp add --transport http --scope project docs https://docs.example.com/mcp

In OpenCode, the same server goes in opencode.json at the project root:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "docs": { "type": "remote", "url": "https://docs.example.com/mcp" }
  }
}

Commit both files so teammates get the same tools, and keep keys in environment variables.

How do I split work between them?#

1. A worktree each. git worktree add ../app-opencode -b opencode/rate-limits gives OpenCode its own folder, so the two agents never edit the same checkout.
2. One owner per task. Write down which agent has which task before either starts.
3. Review across models. Let the agent that did not write a change review its pull request. Different models miss different things.
4. Hand off with a note. Commit, then write the branch, the commit, what is done and what is next. Neither agent can read the other's conversation, so the note is all the receiving one knows.

Handing work from Claude Code to Codex shows a handoff note that works for OpenCode too.

Where does this break down?#

For one person on one computer, this is enough. It strains when several people are involved: a teammate's OpenCode session has no way to learn that your Claude Code session already took the task, a claim written in a file reaches others only after a push and a pull, and a decision made in one session stays there unless someone writes it down.

Where does Joxo fit?#

Joxo joins each teammate's own agent into one project: OpenCode, Claude Code, Codex, Cursor and more, each on its own person's computer and model. Only one agent can hold a task on the board, decisions reach every agent on its next turn, and handoffs carry the branch and the commit. Joxo runs no model, so OpenCode keeps the provider you chose. Joxo for OpenCode and Joxo for Claude Code cover the setup, which is one line pasted into each agent.

Frequently asked questions#

Can OpenCode use my Claude subscription?#

Check Anthropic's current terms before connecting a Claude subscription to a third-party tool. OpenCode works with many providers through API keys, and running Claude Code itself is the supported way to use a Claude plan.

Do I need both CLAUDE.md and AGENTS.md?#

No. One AGENTS.md is enough for both agents. Keep a CLAUDE.md only for Claude-specific instructions, and import AGENTS.md from it so the shared rules still load.

Can OpenCode use my Claude Code skills?#

Yes. OpenCode loads skills from .claude/skills/ in the project and from ~/.claude/skills/, alongside its own folders, so a skill written for Claude Code works in both (OpenCode's skills docs).

Which one should write and which should review?#

There is no fixed answer. Many people let the agent they trust most on the codebase write, and the other review, then swap when one gets stuck. What matters is that the same agent does not write and approve the same change.