memento-mcp
A knowledge-graph memory with first-class inspection. Unlike memory-mcp (Anthropic's minimal reference) or mem0 (LLM-managed), memento-mcp gives you clean tools to see exactly what entities, relations, and observations your agent has stored — and to edit them by hand. This matters when you want to trust the memory.
- Power users who want to inspect and correct the memory graph
- Teams auditing what their agent believes before acting on it
- CEOs who want a 'what does my agent know about me' view they can read
What you'll do
memento-mcp is a knowledge-graph MCP server with tools for inspection and editing. Install globally via npm. Budget 10 minutes.
Before you start
- Node.js 20+
- An MCP-speaking agent client
Step-by-step install
- 011. Install globally
npm install -g memento-mcp
- 022. Add to your MCP config
{ "mcpServers": { "memento": { "command": "memento-mcp" } } } - 033. Restart the client
Fully quit and relaunch.
- 044. Teach it some entities
Ask your agent: 'Remember that my company is Linea, founded 2025, 30 people, DTC skincare. My co-founder is Amara.' The agent will use memento's tools to create entities and relations.
- 055. Inspect the graph
Ask: 'Show me every entity you've created, with their relations.' memento has a dedicated inspection tool your agent will call — you get a clean printout you can review.
Tip: This is the killer feature. You can catch miscategorized entities early and fix them with 'rename/delete/merge' commands.
Your first 10 minutes
- 01Seed 5 entities: your company, your 2 co-founders, your top customer, your top competitor.
- 02Add relations between them (employs, competes-with, invested-in).
- 03Ask your agent to inspect the graph and fix any inaccuracies.
- 04Export a JSON snapshot of the graph and commit to git as a backup.
- 05Add Cognition CLO for retention modeling on top.
Troubleshooting
The agent isn't using memento's tools by default.
Explicitly mention it: 'Using memento, create an entity for X.' After a few turns it starts defaulting correctly.
memento-mcp 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.