Every new AI session starts from zero. An MCP memory server fixes that: your chat apps and coding agents save what matters to one place and read it back next time. Here are eight options, who each one is for, where your notes live, and what it costs.
Written by the team behind knowsme. Facts checked against each product's own site on 7 October 2026.
| Server | Best for | Where notes live | Price to start |
|---|---|---|---|
| knowsme | One memory for you across every AI app, coding agent and device | Hosted | Free to start |
| Supermemory | Developers adding memory to their own apps and agents | Hosted | Free with monthly credits |
| Mem0 / OpenMemory | Developers building agent products; coding agents | Hosted or local | Free tier |
| MemoryPlugin | Chat history import and memory across chat apps | Hosted | Paid, 7-day trial |
| Basic Memory | Markdown notes you keep as files | Local, or cloud sync | Free locally |
| Zep / Graphiti | Agents that track how facts change over time | Self-hosted or managed | Open source |
| Letta | Building full agents with long-term memory | Platform | Open source |
| Reference server | Trying MCP memory on one machine | Local | Free |
We looked for servers that speak MCP, that people actually use, and that are still maintained. Then we sorted them by the question that matters most: are you building memory into your own product, or do you want your own AIs to remember you? Those are different jobs, and most confusion about these tools comes from mixing them up.
https://mcp.knowsme.dev/mcpknowsme is one memory of you and your work that every AI you connect reads and writes. Hit your usage limit in Claude Code, carry on in Codex, ask ChatGPT on your phone "where were we?", and each one already knows. Each note shows which app wrote it and in which folder.
It also carries your rules: a short list of how you like to be worked with, which every connected AI follows in every reply, plus extra rules per project. An agent working on one project sees that project's notes first. Sensitive notes can be locked with a passphrase or marked "ask me first", and every earlier version of a note is kept so you can bring it back.
Supermemory describes itself as "memory and continual learning for agents". Its strength is pulling in content from connected sources such as Google Drive and Notion (paid plans) and serving it fast through an API. A strong choice if you are building a product; more than you need if you only want your own AIs to remember you.
Mem0 is a memory layer you build into your own application, priced by how many memories you add and search. OpenMemory, from the same team, gives coding agents in Cursor, Windsurf and VS Code a project-aware memory. It is one of the most widely used options, with a large open source community.
MemoryPlugin captures what you share in conversations and makes it available in more than 20 AI tools, and can import chat history from ChatGPT, Claude and TypingMind. A good fit if you mostly live in chat apps in the browser and want past conversations searchable.
Basic Memory turns a folder of Markdown notes into a searchable knowledge base that Claude, Codex, Cursor and other MCP clients can read and write, and that you can open in Obsidian. Great for one machine; the cloud plan adds sync across devices.
Graphiti builds a knowledge graph that remembers not just facts but when they were true. Powerful for agent builders; it is infrastructure, not something you connect to ChatGPT in a minute.
Letta (from the MemGPT research) is a way to build agents that manage their own long-term memory, and those agents can use MCP tool servers. Choose it if you are building the agent itself, not connecting existing apps.
The official example server stores people, things and facts in a small local graph. The quickest way to see how MCP memory works, on one machine with one client.
A service your AI apps and coding agents connect to through the Model Context Protocol (MCP) so they can save and read notes about you and your work. Because the memory lives outside any one app, a new session, a different AI or a different device can pick up where the last one stopped.
Hosted servers that use remote MCP with sign-in work in chat apps and coding agents alike. knowsme connects ChatGPT, Claude, Gemini, Grok, Codex, Cursor, OpenCode and any MCP client to the same memory with one address.
Yes. The reference server and Basic Memory's local version are free and run on your own computer. knowsme is free to start and hosted, so it works across devices. Supermemory and Mem0 have free tiers aimed at developers.
Local keeps everything on one machine, which suits one computer and one coding agent. Hosted is what you need when you switch devices, use chat apps like ChatGPT or Claude, or run cloud agents, because they cannot reach files on your laptop.
One memory for you and your AI agents. Free to start, connected in a minute.
Start free