Memory MCP server for Claude and ChatGPT: how it works and what to check
In short
How a memory MCP server gives Claude and ChatGPT one shared memory: five tools, filing by project, and five things to check before you trust one.
By FezolaPublished 2 min read

On this page
A memory MCP server gives your AI assistants one memory that lives outside any of them: whatever one assistant saves, another connected assistant can recall. Instead of each assistant keeping its own slice of what you told it, the server stores your preferences, people, projects and decisions, and serves them through a few tools. This article explains how Fezola's memory server works, then lists five things to check before you trust any memory server with your information. For the comparison with the memories built into Claude and ChatGPT, see Claude memory vs ChatGPT memory.
The five tools
Fezola's memory tools are served to any assistant connected to Fezola at https://api.fezola.com/mcp:
- remember saves one or several memories at once (up to 20 per call), each a short, self-contained statement of up to 2,000 characters, with up to 10 tags and a category.
- recall searches the memory by topic, project or person.
- update memory changes a memory, or moves it to another project.
- forget deletes a memory for good.
- list memories browses everything, page by page.
Filed by project
Each memory belongs to a project, such as a client or an activity, or to a common base called General, for who you are and what you prefer. An assistant working on one project reads that project plus General. In the Fezola app you can limit an assistant to one project, so a work assistant never sees your personal memories. Secrets such as passwords, API keys and verification codes are refused.
Five things to check before you trust a memory server
- Where is it stored, and who can read it? Fezola's memory belongs to your account and is visible to your assistants only through your connector.
- Is it encrypted? Fezola's memory is encrypted at rest. It is not end-to-end encrypted, because the server holds the key.
- Can you see, edit, delete and export it? In the Fezola app you can browse, search, edit, delete one memory or erase all, export, or turn Memory off.
- Are secrets refused? A memory server should never keep passwords or keys.
- How does it handle hidden instructions? A memory is data, not an instruction. Fezola treats memories that way, and a save requested by an email or a document comes from its author, not from you. Keep your assistant's approval prompts on, and you can turn on a confirmation for saves in the settings.
How to connect it
Add Fezola to Claude (see how to add a custom connector to Claude), and to ChatGPT on Business, Enterprise and Edu plans, which allow custom connectors. Memory is on by default and the app lets you turn it off, assistant by assistant. We have verified the memory with Claude; ChatGPT receives the same tools.
Read the full picture in one shared memory for Claude, ChatGPT and all your AI assistants, and see the other tools you can connect in best MCP servers for Claude and ChatGPT. New to MCP? Start with what is MCP.
Questions and answers
What is a memory MCP server?
It is an MCP server that stores facts about you and your work and serves them to AI assistants through tools such as remember and recall. Because the memory lives in the server, not in one assistant, every connected assistant can use it.
Which tools does Fezola's memory server offer?
Five: remember, recall, update memory, forget and list memories. Remember can save up to 20 memories at once, each a short statement of up to 2,000 characters with up to 10 tags.
How are memories organized?
By project. Each memory belongs to a project, such as a client or an activity, or to a common base called General for who you are and what you prefer. An assistant working on one project sees that project plus General, and you can limit an assistant to one project.
What should I check before using a memory MCP server?
Where the memory is stored and who can read it, whether it is encrypted, whether you can view, edit, delete and export it, whether secrets are refused, and how it handles instructions hidden in emails or pages.
Can an email poison the memory?
Fezola serves memories to the assistant as data, never as instructions, and tells it that a request found in an email or document comes from its author, not from you. You can also turn on Ask for my confirmation in the settings.
Is it free?
Fezola is free with 2 sources, and Memory is not a source. Pro connects unlimited sources, with a 7-day free trial.
More in Memory and privacy

Claude memory vs ChatGPT memory: how to share one memory across both
How Claude's and ChatGPT's memories work, why each keeps its own slice, and how Fezola Memory gives both one shared, encrypted memory you control.
2 min read

One shared memory for Claude, ChatGPT and all your AI assistants
Fezola Memory lets every assistant recall what you told another one. How it works, how it is filed by project, and how to view, edit or erase it.
3 min read

Is it safe to connect an MCP server to Claude or ChatGPT? A checklist
Connect only trusted MCP servers, check permissions, limit tools, keep approvals on and know how to revoke. A checklist from official sources, Nov 2026.
3 min read