How to Migrate from Claude to OpenClaw

OpenClaw has a Claude migration provider for people coming from Claude Code or Claude Desktop. The docs emphasize preview and reversibility: every item is previewed before state changes, secrets are redacted in plans and reports, and a verified backup is created before apply.

30-second answer

Use the onboarding import flow for a fresh OpenClaw setup, or openclaw migrate for scripted runs. The importer can move Claude instructions, memory, MCP server definitions, skills, and command Markdown files into the OpenClaw workspace shape. Review the dry run before applying.

Onboarding path

openclaw onboard --flow import
openclaw onboard --import-from claude --import-source ~/.claude

The onboarding wizard offers Claude when it detects local Claude state. This is the friendliest path for a clean machine or first OpenClaw setup. The docs note that onboarding imports require fresh OpenClaw state; if local state already exists, reset config, credentials, sessions, and workspace first or use the migrate CLI carefully.

CLI path

openclaw migrate claude --dry-run
openclaw migrate apply claude --yes

Add --from <path> when importing a specific Claude Code home or project root. The dry run is the important step. It lets you see the plan, redactions, and backup behavior before OpenClaw touches the destination.

What gets imported

Project CLAUDE.md and .claude/CLAUDE.md content is copied or appended into the OpenClaw agent workspace AGENTS.md. User ~/.claude/CLAUDE.md content is appended into workspace USER.md. MCP server definitions can be imported from project .mcp.json, Claude Code ~/.claude.json, and Claude Desktop config when present.

Claude skills with a SKILL.md file are copied into the OpenClaw workspace skills directory. Claude command Markdown files under command roots are imported as well, preserving useful operator workflows without pretending every Claude-specific behavior has the same runtime semantics.

Migration cautions

Do not import blindly into a working OpenClaw home. Old Claude instructions may contain assumptions that conflict with OpenClaw channels, tools, memory, approvals, or safety rules. After import, read the workspace files, remove stale commands, and test one harmless session before wiring public channels.

The OpenClaw Playbook turns this into a clean migration checklist: dry-run, backup, inspect imported identity, verify MCP servers, test skills, then only connect business channels after the agent behaves correctly in a private session.

Rollout plan

Treat How to Migrate from Claude to OpenClaw as a workflow you roll out in stages, not a switch you flip once. Start with the smallest harmless proof: a status check, dry run, local-only call, private session, or read-only inspection. Confirm the documented behavior matches your installed OpenClaw version, then write the exact commands and expected output into the workspace so the next agent does not rely on memory or vibes.

For a production runbook, document installation state, service ownership, update cadence, rollback command, and the exact machine that owns the Gateway. Also write down what the agent may do alone, what requires approval, and what must stop immediately. That boundary is the difference between useful autonomy and a workflow that surprises the operator at the worst possible time.

Keep one rollback note beside the guide. It can be as simple as the command to disable a plugin, the channel to pause, the config key to revert, or the owner who must approve the next run. Include the proof that tells you rollback worked, and keep it visible near the production checklist for future maintainers. Agents are most useful when recovery is obvious.

After the first live run, review the transcript or logs while the details are fresh. Look for missing prerequisites, stale assumptions, broad prompts, confusing errors, and any external side effect that should have been gated. Tighten the guide, then repeat with one wider scope. The OpenClaw Playbook is built around this operating rhythm: cautious first proof, written runbook, verified automation, then gradual autonomy once the evidence is boring.