agentmemory
The 'zero-config' memory layer for agent-first developers. Install it globally, add one line to your MCP config, and it works the same across Claude Code, Cursor, Gemini CLI, and OpenCode. It silently captures your sessions and injects the relevant context when you start the next one.
- Developers who switch between multiple AI clients
- Anyone who wants one memory layer shared across all their agents
- Zero-config purists — this one's close
What you'll do
agentmemory installs globally via npm and registers as an MCP server for every compatible client. Budget 5 minutes.
Before you start
- Node.js 20+
- At least one MCP-speaking agent client
Step-by-step install
- 011. Install globally
One npm command.
npm install -g agentmemory
- 022. Add to your MCP config
Paste into each client's .mcp.json.
{ "mcpServers": { "agentmemory": { "command": "agentmemory" } } } - 033. Restart your clients
Fully quit and relaunch each agent. agentmemory starts capturing on the next session.
Your first 10 minutes
- 01Use Claude Code for 30 minutes. Start a new session.
- 02Ask 'what do you remember from last session?' Confirm retrieval works.
- 03Switch to Cursor. Same memory should be accessible there.
- 04Review the memory log (agentmemory ships a CLI: `agentmemory list`).
- 05Add Cognition CLO for the retention layer.
Troubleshooting
`agentmemory` command not found after install.
Check your global npm bin path is on PATH. `npm config get prefix` shows where globals go; add its `bin` directory to your shell's PATH.
Memories aren't carrying across clients.
Confirm both clients have the agentmemory entry in their MCP config. Each client reads its own config — it doesn't auto-propagate.
agentmemory holds the knowledge. Cognition CLO models retention per employee per concept using a Weibull forgetting curve — so you see decay before it becomes a missed SOP or a failed audit.