agentmemory: one memory layer that works across Claude Code, Cursor, Gemini CLI, and OpenCode
A plain-English guide to agentmemory — the zero-config MCP memory layer that gives your multiple AI clients one shared brain. 5-minute install. Built for developers who switch between agent tools.
Short version: If you use more than one AI coding agent — Claude Code in the morning, Cursor in the afternoon, Gemini CLI for scripts — agentmemory gives all of them one shared memory. Install globally via npm, paste one line into each client's MCP config, and your agents stop starting from zero every time you switch. 5-minute install. By rohitg00, MIT licensed.
What is agentmemory?
agentmemory is the "zero-config" memory layer for developers who work across multiple AI agent clients. It installs globally with npm install -g agentmemory, registers as an MCP server, and the same memory is accessible from every compatible client: Claude Code, Cursor, Gemini CLI, OpenCode, and anything else that speaks MCP.
The value prop is agent portability. If you've ever written a great CLAUDE.md only to switch to Cursor and lose all that context — agentmemory fixes that. One memory, many surfaces.
Who this is for
- Developers who genuinely use more than one AI agent (Claude Code + Cursor is the common pair).
- Anyone who wants to try new agents (Gemini CLI, OpenCode) without losing their existing memory setup.
- Zero-config purists — this one is close. Literally
npm i -gand one config line. - Teams standardizing on MCP but not on a single client.
Skip this if
You only use one agent and want fine-grained memory control. For single-client setups with opinionated memory, look at mcp-memory-service (Claude Code-tuned) or memento-mcp (inspectable graph).
What problem it solves
AI agent memory is typically locked to the client that wrote it. Claude Code has its memory. Cursor has its memory. Your agents don't talk to each other. Every time you switch clients — or try a new one — you re-teach.
agentmemory unifies this. The memory layer lives outside any single client. Every agent reads and writes the same store. You switch tools without cognitive cost; your agents act like they've been collaborating all along.
How to install it (plain English)
- Install globally.
npm install -g agentmemory. One command, global install. - Add to your MCP config. In every client's
.mcp.json:{ "mcpServers": { "agentmemory": { "command": "agentmemory" } } } - Restart your clients. Fully quit and relaunch each one.
agentmemorystarts capturing on the next session.
That's it. Full walkthrough: /memory/tools/agentmemory.
What you can do with it (for a non-technical founder)
If your team uses multiple AI clients (common as companies hedge between Claude, Cursor, and Gemini):
- One shared brain across tools — whatever Claude Code learns Monday, Cursor knows Tuesday.
- Faster onboarding for new agents — evaluating a new tool doesn't mean starting fresh. Your memory travels.
- Less "re-explaining" overhead — one-time teach, all-clients remembers.
- Client-agnostic memory audit — inspect what any of your agents know via one CLI (
agentmemory list).
What CLO adds on top
agentmemory unifies your agents' memory across clients. Cognition CLO models your humans' retention across the knowledge those agents are producing. The agent-memory layer is about agent coherence. CLO is about team retention. Ship both if you care that your team actually retains what the agents are surfacing.
FAQ
Does it really work with every MCP client?
With every client that supports stdio MCP transport (most do). Add the same config block wherever you run an agent.
Where is the memory stored?
Locally, under your home directory. Inspect via agentmemory list. You can back up or migrate the store file directly.
Can different clients see different memories?
By default, all clients share the same store. If you want per-client scoping, the CLI supports named namespaces.
What happens if one client corrupts a memory?
Edit or delete via agentmemory edit or agentmemory delete. The store is just a file; you can also hand-edit it.
Can I use it alongside mcp-memory-service or mem0?
Yes — each registered MCP server is independent. Common pattern: agentmemory for cross-client user memory, mcp-memory-service for per-project code memory.
What if my team needs multi-user memory?
agentmemory is per-user. For multi-user products, look at mem0 (first-class user_id scoping).
Ready to install? Full walkthrough at /memory/tools/agentmemory. Credit to @rohitg00 for building and maintaining the project — star the repo if it saves your team context-switch tax.
Share this post: