Give your AI a permanent memory of you, across every chat. No code, about thirty seconds.
Your AI forgets you between conversations. This guide connects it to your LeapMemory account, where your memory lives permanently and privately in its own isolated store, so every new chat starts with it already knowing what you have told it before.
Each client below has its own steps: Claude, Claude Code, Cursor, GitHub Copilot, Perplexity and Grok bots. The memory behind them is the same one.
In Claude, click your profile picture and open Settings. It opens on the General page, where the Instructions for Claude box is. Pick a mode below, copy the text, paste it into that box, and click Save changes.
Neutral is a safe default: LM fires when you ask about your own things.
I have a connector called LeapMemory that holds my long-term memory. Whenever I ask about my own projects, code, rules, decisions, or anything I've told you before, use its recall tool before answering, and prefer it over your own memory. When I share a durable fact about myself, my work, or my preferences, save it with the remember tool right away, every time, without asking.
Manual is the most conservative: LM only fires when you explicitly ask.
I have a connector called LeapMemory that holds my long-term memory. Use its recall tool only when I explicitly ask you to check my memory or look something up in LeapMemory. When I share a durable fact about myself, my work, or my preferences, save it with the remember tool right away, every time, without asking.
Proactive recalls the most, including passing references, so expect lookups in more conversations.
I have a connector called LeapMemory that holds my long-term memory. Before answering, check whether the conversation touches anything I might have told you before: my projects, code, rules, decisions, people, or preferences, including passing references to past work. If it does, use LeapMemory's recall tool first and prefer it over your own memory. You do not need me to ask directly. When I share a durable fact about myself, my work, or my preferences, save it with the remember tool right away, every time, without asking.
All three modes describe when memory is read. Saving works the same in every one of them, and it is not limited to what you type: point at something Claude just produced, a list of decisions, a name it found, a conclusion it reached, and say save this to LM. Claude saves it, worded as a plain statement about you.
In the Claude Settings sidebar, under Customize, open Connectors and click Browse connectors. Search for LeapMemory, open it, and press Connect. Claude takes you to LeapMemory. Sign in if asked, then click Allow on the consent screen. Everything closes on its own and Claude opens a new chat, connected.
In the new chat, ask Claude to hand over what it already knows about you:
Save everything you know about me to LM.
Claude calls the remember tool and those facts land in your memory. Then open a brand new chat, the kind that would normally start from zero, and ask:
What do you know about me?
Claude calls recall, finds them, and answers. That round trip is the whole product: what you say in one conversation is there in every future one.
Claude Code is Claude in your terminal, and it carries your memory too. It is the same account and the same isolated store: what you save in the chat window is there in the terminal, and the other way around. The three steps above translate directly.
First check whether the connector is already there. If Claude Code is signed in with the same Claude account, connectors you added in Claude on the web show up in Claude Code automatically:
claude mcp list
If LeapMemory is listed, skip the next command. Otherwise add it, one command instead of a Settings page:
claude mcp add --transport http --scope user leapmemory https://api.leapmemory.com/mcp
If you ran that command, open a Claude Code session, run /mcp, choose leapmemory, sign in if asked, and click Allow in the browser page that opens. If LeapMemory was already listed, it is already approved and there is nothing to do here. It is the same approval as step 2, and --scope user makes the connector available in all your projects.
The instruction lives in a file instead of a Settings box. Add a ## Memory section to ~/.claude/CLAUDE.md, Claude Code's personal memory file, loaded at the start of every session, and paste your preset under it. The Connect page in the app has a Claude Code version of each preset, worded for codebases: decisions, conventions, and root causes.
The test is identical: save in one session, open a fresh one, and ask what it knows.
Every client connects to the one URL below, over OAuth: there is no API key to paste. You sign in to LeapMemory in your browser and press Allow on the consent screen, which reads “<client> wants access to your LeapMemory”.
https://api.leapmemory.com/mcp
Cursor reads its servers from a file. Create ~/.cursor/mcp.json with exactly this, or add the leapmemory entry to the file you already have. No type field and no headers are needed. For one project only, the same file works at .cursor/mcp.json inside the project.
{
"mcpServers": {
"leapmemory": {
"url": "https://api.leapmemory.com/mcp"
}
}
}
Then open Customize in Cursor’s sidebar. Under MCPs, leapmemory appears with a User tag and the words Needs Authentication in Local. Click Authenticate on the right, approve the consent screen in your browser, and come back. Clicking leapmemory should now show Local: Connected and the six tools: recall, briefing and list_rules under Reads, remember, add_rule and update_rule under Writes.
Open the Command Palette, run MCP: Add Server, choose HTTP, and give it the server URL. Name it anything; VS Code’s suggested name works. When VS Code asks where to save it, pick User Configuration. Then press Start above the server entry in mcp.json and approve twice: once in VS Code, once on the LeapMemory consent screen.
Perplexity Computer has no “add MCP connector” field on the free plan, so LeapMemory goes in as a small personal plugin instead. Open Plugins in the sidebar, click Add custom, choose Upload plugin or skill, and upload this plugin zip. Upload the zip itself, do not extract it.
The zip holds nothing but these two files, if you would rather build it yourself:
{
"name": "leap-memory",
"description": "Connect Perplexity Computer to the LeapMemory remote MCP server."
}
{
"mcpServers": {
"leap-memory": {
"type": "streamable-http",
"url": "https://api.leapmemory.com/mcp"
}
}
}
Then press Connect on the plugin card and approve the consent screen.
plugin.json must be leap-memory, lowercase with a hyphen. Perplexity rejects Leap Memory with “use only lowercase letters, numbers, and hyphens in the name”.Ask the bot in plain words, for example “Connect LeapMemory. The MCP server is api.leapmemory.com/mcp”. Grok adds it as a custom connector and shows a LeapMemory card; press Sign in on that card and approve.
The connector speaks MCP, the open protocol AI assistants use to reach external tools, and the connection is authorized with OAuth: your AI never sees your password, and you can revoke access at any time. Once connected, your client gets six tools. Three carry your memory: recall, which searches it, briefing, which loads your background picture when a conversation starts, and remember, which saves to it. Three manage your standing rules, instructions Claude follows in every conversation: add_rule saves one, list_rules shows them, and update_rule changes one. In the LeapMemory app you can view and delete your memories, and view, edit, and delete your rules. To correct a memory, tell your AI plainly that the old version is wrong and what is right instead, or delete it in the app.
Your memory itself lives in your own physically isolated store, separate from every other person's, the same isolation architecture the developer API provides. Recall is free and unlimited on every plan; saving counts toward your plan's memories.
Claude never uses LM. Check both halves of the setup: the instruction is actually saved in the Instructions for Claude box (step 1), and the connector shows as connected under Settings, Connectors (step 2). The connector without the instruction sits unused; the instruction without the connector has nothing to call.
Recall misses something you know is saved. Recall works best with specific words: names, projects, terms you actually used. "What did I say about the Atlas deploy" finds more than "what did I say recently."
A just-saved fact does not come back yet. Memories form in the background and typically land within seconds. If you save and immediately recall in the same breath, give it a moment and ask again.