One CLI, every agent. Install once and memoryclaw setup writes the skill block into the right file for whatever agent (or agents) you use — Claude Code, Cursor, Aider, OpenClaw, and more.
Install once, works everywhere
1. Install
One curl installs the memoryclaw CLI globally on macOS, Linux, and WSL.
2. Setup
memoryclaw setup signs you in, picks a passphrase, and detects every supported agent on your machine.
3. Paste or auto-wire
The skill block is written between idempotent sentinel markers. Prefer manual? Run memoryclaw setup --print and paste into any rules file.
Default install ships in memory mode (Neural Memory only). Want backup too? Set MEMORYCLAW_MODE=full — works for any agent (Claude Code, Cursor, Codex, OpenClaw, …).
After install — what to expect, how to verify
Setup writes a small skill block into each agent's instruction file. Agents read it on next session start — there's no in-session reload for most of them. Quick verification:
Reload your agent
Claude Code: quit (Ctrl+C) and re-run claude
Cursor: ⌘⇧P → “Developer: Reload Window”
Aider / Codex CLI / Goose: quit and re-run the CLI
Gemini CLI: quit and re-run gemini
OpenClaw: restart your OpenClaw session — the plugin reads ~/.openclaw/workspace/AGENTS.md on launch
Verify it's wired
$
memoryclaw doctor --agents
Lists each agent's sentinel block + the Claude Code Skill plugin status. ✓ means the file landed; ⚠ means re-run setup for that target.
$
memoryclaw memory recall "hello" --thread all
Inside the agent or your terminal — confirms the CLI is authenticated and Neural Memory is reachable.
Want a visible Skill in Claude Code's /skills?
The CLAUDE.md inject above is passive context Claude reads on session start. Optional: also install MemoryClaw as a native Claude Code Skill so it appears in the /skills palette and is invokable directly:
$
claude plugin install ngsrv/memoryclaw
Per-agent guides
Run the install above first, then pick your agent. Have several? Run memoryclaw setup with no flags — it detects all of them at once.
Claude Code
Anthropic's CLI agent. Two layers: passive context in ~/.claude/CLAUDE.md, plus an optional visible Skill in /skills.
Setup command
$
memoryclaw setup --for claude-code
Skill block lands in
~/.claude/CLAUDE.md (+ optional: claude plugin install ngsrv/memoryclaw)
After setup runs the CLAUDE.md inject, run `claude plugin install ngsrv/memoryclaw` to also get MemoryClaw as a visible Skill in the /skills UI. Verify with `memoryclaw doctor --agents`.
Cursor
Cursor reads project AGENTS.md and .cursor/rules/.
Setup command
$
memoryclaw setup --for agents-md
Skill block lands in
AGENTS.md (project-level) or ~/AGENTS.md (user-level)
User-level by default. Pass --project inside a git repo to write a project-scoped block instead.
OpenClaw
Plugin loader. Backs up ~/.openclaw/ — alongside any other agents you opt into.
Setup command
$
openclaw plugins install clawhub:memoryclaw
Skill block lands in
~/.openclaw/plugins/memoryclaw/
OpenClaw users get the full install (Neural Memory + encrypted backup). Other agents (Claude Code, Cursor, Codex, Gemini, Aider, Goose) also support encrypted backup — see /docs.
Aider
AI pair programmer. Reads AGENTS.md and CONVENTIONS.md.
Setup command
$
memoryclaw setup --for agents-md
Skill block lands in
AGENTS.md (or CONVENTIONS.md)
OpenAI Codex CLI
OpenAI's coding CLI. Reads AGENTS.md natively.
Setup command
$
memoryclaw setup --for agents-md
Skill block lands in
AGENTS.md
Gemini CLI
Google's CLI. ~/.gemini/GEMINI.md takes precedence over AGENTS.md.
Setup command
$
memoryclaw setup --for gemini
Skill block lands in
~/.gemini/GEMINI.md
Goose
Block's autonomous developer agent. Reads AGENTS.md and .goosehints.
Setup command
$
memoryclaw setup --for agents-md
Skill block lands in
AGENTS.md (or .goosehints)
Cline
VS Code agent. Project-level rules.
Setup command
$
memoryclaw setup --print
Skill block lands in
Manual paste into your Cline rules file
No automated wiring yet — copy the block from --print and paste into your Cline rules.
Continue
Open-source IDE assistant. Project-level rules.
Setup command
$
memoryclaw setup --print
Skill block lands in
Manual paste into your Continue config
No automated wiring yet — copy the block from --print and paste into your Continue config.
Claude Desktop
MCP-only. No CLAUDE.md — wire via the MCP server.
MCP — Phase 2
Setup command
$
memoryclaw mcp serve
Skill block lands in
Claude Desktop MCP config
MCP server lands in Phase 2. Until then, Claude Desktop users can drive memory through the CLI in a terminal alongside Claude Desktop.
Custom / REST
LangChain, OpenAI SDK, your own scripts — anything that can shell out or call REST. Each saved memory is called an "engram" in the CLI and API.
Setup command
$
memoryclaw memory engram --auto --message "…"
Skill block lands in
(no skill file — your code calls the CLI directly)
REST endpoints under /api/cortex are documented in /docs#neural-memory. Programmatic SDK is on the roadmap.
Don't see your agent?
If your agent reads any of AGENTS.md, CONVENTIONS.md, or a project-level rules file, the universal setup target works:
$
memoryclaw setup --for agents-md
Otherwise, run memoryclaw setup --print to copy the canonical skill block, then paste it into whatever instruction file your agent reads. Or call the memoryclaw memory * commands directly from your agent code — the CLI is the universal interface.
Need the full reference? Read the docs, browse pricing, or jump straight into your dashboard.