OpenCode vs Claude Code: open source agent vs Anthropic's
OpenCode vs Claude Code from their docs: open source and 200+ providers vs Claude models, AGENTS.md vs CLAUDE.md, skills, surfaces, and how to share a setup.
A team wants one coding agent across developers who use different model providers: some on Claude, some on GPT, a few on local models. Claude Code runs Claude models. OpenCode runs almost anything. The question is what else differs, and whether the two can share a setup.
This guide compares them from OpenCode's docs and Claude Code's, checked on 16 November 2026. It leaves out pricing and benchmarks.
The short answer
| OpenCode | Claude Code | |
|---|---|---|
| License | Open source, MIT | Anthropic product |
| Models | Free models or 200+ providers | Claude |
| Surfaces | Desktop, terminal, mobile, SDK | Terminal, IDEs, desktop app, web, mobile, Agent SDK |
| Instruction file | AGENTS.md only | CLAUDE.md (can import AGENTS.md) |
| Project skills | .opencode/skills, plus .claude/skills and .agents/skills | .claude/skills |
| MCP | Yes | Yes |
| Permissions | Yes | Yes, plus auto mode with a classifier |
| Org policy | Policies, deliverable from an OpenCode Console workspace | Managed settings |
OpenCode
OpenCode describes itself as the open source coding agent. Its site lists a desktop app with sessions, tabs, diffs, and a built-in browser on macOS, Windows, and Linux; a terminal UI that works in any shell or on a remote server; a mobile app to start, steer, and approve sessions while the agent runs on your computer; and an SDK for scripts, CI, and tools. The current major version is V2.
Instructions. OpenCode reads AGENTS.md: project files are committed and shared, a global file applies everywhere, and files deeper in the tree load as the agent explores. Its docs are explicit that V2 recognizes AGENTS.md only and doesn't fall back to CLAUDE.md.
Skills. Skills live in .opencode/skills/<id>/SKILL.md. For compatibility, OpenCode also reads .claude/skills and .agents/skills in the project, and ~/.claude/skills and ~/.agents/skills globally.
Claude Code
Claude Code is Anthropic's agentic coding tool, running Claude models across the terminal, IDE extensions, the desktop app's Code tab, and cloud sessions on the web and mobile. Its distinguishing features are mostly about teams and scale: hooks at many lifecycle points, permission modes including auto mode, managed settings for organization policy, agent teams, workflows, and cloud sessions.
Sharing one setup
The two can share most of a repository's configuration:
- Instructions in
AGENTS.md. OpenCode reads it directly. InCLAUDE.md, add@AGENTS.mdto import it, plus any Claude-specific notes. See AGENTS.md examples. - Skills in
.claude/skills/. Both read it. Stick to the Agent Skills spec fields so both parse them. - MCP servers configured in each tool, pointing at the same servers.
Which to choose
- You need many model providers, local models, or open source: OpenCode.
- You're standardizing on Claude and want organization policy, hooks, and cloud sessions: Claude Code.
- Mixed team: shared
AGENTS.mdand.claude/skills, each person on the agent that fits.
What both are missing
Both agents read the repository's instructions. Neither knows a decision the team made last week unless it's written where they look, and instruction files aren't built for decisions that change and need an owner.
Where Lockstep fits
Lockstep captures decisions from PRDs, Slack, and pull requests, routes each to its owner for approval, and briefs Claude Code sessions automatically. Briefings, hooks, and managed skills are Claude Code features; OpenCode and other MCP clients can call Lockstep's tools directly.
See how it works.
Sources checked 16 November 2026
- https://opencode.ai/
- https://opencode.ai/docs/instructions/
- https://opencode.ai/docs/skills/
- https://opencode.ai/docs/policies/
- https://github.com/anomalyco/opencode
- https://code.claude.com/docs/en/overview
Related reading
For other comparisons, see Gemini CLI vs Claude Code vs Codex CLI and Cursor vs Antigravity vs Claude Code. For all the alternatives, see Claude Code alternatives.
Frequently asked questions
- What is OpenCode?
- An open source coding agent, released under the MIT License, that you can use from a desktop app, the terminal, your phone, or your own code through an SDK. Its site says you can start with free models or bring any of 200+ providers.
- What is the difference between OpenCode and Claude Code?
- OpenCode is open source and model-agnostic, working with many providers. Claude Code is Anthropic's agent and runs Claude models, with Anthropic-specific features such as managed settings for organizations, auto mode, agent teams, and cloud sessions. They use different instruction files: AGENTS.md for OpenCode, CLAUDE.md for Claude Code.
- Does OpenCode read CLAUDE.md?
- No. OpenCode's docs say V2 recognizes AGENTS.md only and doesn't use CLAUDE.md as a fallback. To share instructions, keep them in AGENTS.md and import it into CLAUDE.md with @AGENTS.md.
- Can OpenCode use Claude Code skills?
- Yes. OpenCode's docs list .claude/skills and .agents/skills as compatibility locations for project skills, and ~/.claude/skills and ~/.agents/skills for global ones, alongside its own .opencode/skills.
- Can OpenCode use Claude models?
- OpenCode connects to many model providers through its providers configuration. Check its providers docs for current options and any terms that apply to the provider and account you plan to use.
Keep reading