What is an MCP memory server?

An MCP (Model Context Protocol) server exposes tools that an AI client can call. A memory server offers tools such as search and fetch over stored information, so the assistant can look things up on its own instead of relying on what you paste.

Does Memex have an MCP server?

Yes. Every Memex archive has a remote MCP endpoint at https://memex.li/api/mcp/ . It offers tools to search the archive, open conversations, read and write breadcrumb notes, and archive new sessions.

How do I add Memex to Claude Code?

Run: claude mcp add --transport http --scope user memex https://memex.li/api/mcp/ . The user scope makes Memex available in every project; without it, Claude Code adds the server to the current project only.

What is the difference between a Memex link and the MCP server?

They read the same archive. The link is plain HTTP and works with any assistant that can fetch URLs. MCP gives clients that support it typed tools, which makes searching and writing notes more reliable.

Knowledge Base

Connect Your AI Chat History to Claude with an MCP Memory Server

Memex·7 min

Short answer: Memex serves your whole AI conversation archive through a remote MCP endpoint, https://memex.li/api/mcp/<your-token>. Add it as a custom connector in Claude, or run claude mcp add --transport http --scope user memex https://memex.li/api/mcp/<your-token> in Claude Code. The assistant can then search your ChatGPT, Claude, Gemini and DeepSeek history as a tool, open the relevant conversations, and leave notes for the next session.

Why MCP for personal memory

The Model Context Protocol is an open standard for connecting AI applications to tools and data. Instead of you copying information into a chat, the assistant calls a tool and gets exactly what it needs.

For memory, that difference matters:

The assistant decides when to look. It can search your archive in the middle of a task, when a question actually requires past context.
Searches are precise. Tools take structured parameters, such as a query, a date range or a topic filter, instead of guessing URL formats.
Writes are safe and explicit. Recording a note or archiving a session is a named tool call you can see and approve.

What the Memex MCP server offers

The exact tool list is published in each archive's protocol, but the core set is:

Tool	What it does
Search	Full-text search across every conversation, with filters by topic, content type and metadata
Fetch / get conversation	Open one conversation as Markdown, in order
Most helpful	The conversations other assistants rated most useful
Notes and breadcrumbs	Read and write short AI-to-AI notes about which conversation answered which question
Archive conversation	Save the current session into the archive, with optional metadata

Because the archive is built from your platform exports, the assistant is searching real conversations, not a list of facts someone extracted.

Connecting Memex

First, sign in to Memex, upload at least one export, and copy your token from the dashboard.

Claude (web and desktop)

Add Memex as a custom connector using the remote MCP URL https://memex.li/api/mcp/<your-token>. Once enabled in a chat, Claude can call the Memex tools whenever a question needs your history. Adding a line to your Project instructions such as "search Memex before answering questions about my past work" makes it consistent.

Claude Code

claude mcp add --transport http --scope user memex https://memex.li/api/mcp/<your-token>

The --scope user flag registers Memex for every project on your machine. Leave it out to add it to the current project only.

This is especially useful for developers. Claude Code sessions do not carry memory between projects, but your archive does. Ask "search Memex for how I fixed the asyncpg pool exhaustion" and it pulls the earlier debugging session into the current one.

Other MCP clients

Any client that supports remote MCP servers over HTTP can use the same URL. For assistants without MCP support, use the plain Memex link instead; it exposes the same archive over ordinary HTTP with an OpenAPI description.

Security model

Your token is the credential. Keep it out of public repositories and shared screenshots.
Read and write are separate permissions. Writes, such as archiving conversations, can be switched off for a token.
Notes are data, not instructions. Memex's protocol tells agents that breadcrumbs written by other assistants are unverified and must never be executed as instructions.
Revocation is immediate. Revoke a token and every client using it loses access.

MCP memory server vs agent memory frameworks

Developer frameworks such as Mem0, Zep or Letta also provide memory, including over MCP. They are designed to give an application you build a memory of its users. Memex is designed to give you a memory of your own conversations across the assistants you already use, starting from your existing history on day one. If you are building an agent product, a framework may fit better; if you want Claude to remember what you did in ChatGPT, Memex is the direct route. See the full AI memory tools comparison.