MemoryLabs
Setup

Connect once. Carry it everywhere.

MemoryLabs is one connector (the Model Context Protocol) that every major assistant already understands. Add it once per app and the same memory reads and writes everywhere you work, on every device.

Terminal tools — one command does it all
npx memorylabs connect

Run that in a terminal and it finds Claude Code, Codex, Cursor, and Gemini CLI on your machine, wires each one to your vault, and walks you through sign-in. Every file it touches is backed up first. Browser apps connect with the URL below.

Your connector URL (for browser apps)
https://memorylabs.dev/api/mcp
01

Claude (web & desktop)

  1. 1.Open Settings → Connectors and choose Add custom connector.
  2. 2.Paste your connector URL and continue.
  3. 3.Sign in with your email code and approve access. Claude can now read and write your vault.
02

ChatGPT

  1. 1.In Settings → Connectors (developer mode), add a custom connector.
  2. 2.Paste the connector URL and complete the same email sign-in and consent.
  3. 3.Your memory is now shared with ChatGPT, alongside every other assistant you connect.
03

Cursor & IDE assistants

  1. 1.Add the connector as an MCP server in your editor's AI settings.
  2. 2.Authorize once; your coding assistant now remembers your stack and preferences across projects.
04

Gemini, Grok & the rest

  1. 1.Anywhere that accepts a custom MCP connector, paste the same URL.
  2. 2.One vault, one identity, every model. Switching assistants no longer means starting over.
05

Codex CLI

  1. 1.Add MemoryLabs to ~/.codex/config.toml:[mcp_servers.memorylabs] url = "https://memorylabs.dev/api/mcp" auth = "oauth"
  2. 2.Run codex mcp login memorylabs and approve access in the browser with your email code.
  3. 3.Restart Codex, then ask it to load your context. Approve the read tools once (choose “always allow” to skip future prompts). It is now reading and writing the same vault as your other assistants.

One line to make it automatic.

Assistants use your memory when you ask, and reliably every time once you paste this into each app’s custom instructions. It tells the assistant to load your context at the start, check before asking, save as it learns, and summarize at the end.

Custom instructions (paste into each assistant)

I use a MemoryLabs memory connector. At the start of any substantive conversation, call load_context to load what you already know about me. Before asking me something I might have already told an assistant, call recall first. Save durable facts, decisions, and preferences with remember as they come up. Near the end of a working session, call checkpoint with a thorough summary of everything we covered.

Confirm it’s working

Three steps to see the loop close.

01

Ask what it knows

  1. 1.After connecting, start a fresh chat and ask: “What do you know about me?”
  2. 2.A connected assistant calls load_context and answers from your vault. A brand-new vault will say it is just getting started, which is correct.
02

Tell it something

  1. 1.Say something durable, for example: “Remember that I prefer concise answers.”
  2. 2.The assistant saves it with remember. Credentials are refused by design, so never paste a secret.
03

Watch it land

  1. 1.Open your vault and check All records and the Timeline.
  2. 2.Your new memory appears within seconds, dated and attributed to the assistant that saved it. That is the whole loop, proven.
Request access →