One memory for Claude, ChatGPT, Cursor and Grok

The problem

You plan a feature in claude.ai, write the code in Cursor, and ask ChatGPT or Grok a question about it later. Each app keeps its own memory, tied to its own account or folder. None of them can read what the others know, so every switch means explaining the project, the decisions and the open work again.

What each app has built in

• Claude (claude.ai, Desktop, phone)Claude can search your past chats on paid plans, and its memory saves topics from your chats and carries them into new conversations. Each project has its own memory. Source
• Claude CodeCLAUDE.md files you write, and auto memory notes Claude saves for itself, load at the start of each session. Auto memory stays on the computer that wrote it. Our Claude Code memory guide covers this in detail. Source
• ChatGPTWhen Memory is on, ChatGPT can use saved memories, past chats and custom instructions, depending on your plan and region. You manage it in Settings, Personalization, Memory. Source
• CursorCursor's docs say models do not keep memory between completions. Rules give lasting context instead: project rules in .cursor/rules, user rules, team rules, and AGENTS.md. Source

All of these are useful. But each one lives inside its own app, so it does not help the others.

The shared fix: one MCP server

MCP (Model Context Protocol) is a standard way for AI apps to connect to outside tools. Claude, ChatGPT, Cursor and Grok can all connect to MCP servers. Source Source Source Source A memory server is one such tool. Connect every app to the same one, and a decision saved from one app is there when you open another.

FlowState is one such server. There are two ways to reach it:

• LocalFree and open source (MIT). FlowState keeps your memory in one file on your computer, and apps on that same computer (Claude Desktop, Claude Code, Cursor) start it themselves. No account needed.
• Remoteclaude.ai, the Claude phone apps, ChatGPT and Grok connect to MCP servers from their own servers, so they need a public https address. Source Source FlowState Relay (in beta) gives the FlowState on your computer a private address without setting up a tunnel, and your data stays on your computer. Hosted (in beta) keeps your memory on our server instead. Both are optional; see pricing.

Full steps are in the setup guide. The MCP config builder writes the config for you.

Set up each app

First install FlowState on your computer. The easy way: paste this into Claude Code or Cursor and let it follow FlowState's own AGENTS.md.

Set up FlowState for me from https://github.com/dialectforge/FlowState by following its AGENTS.md.

By hand, in a terminal (on Windows type python for python3). Keep only the apps you use after --clients. The first setup line shows what would change; the second makes the change and checks that FlowState answers. Then fully quit and reopen those apps.

git clone https://github.com/dialectforge/FlowState.git FlowState
cd FlowState
python3 setup_flowstate.py --clients claude-desktop,claude-code,cursor --schedule-backups --no-offsite --install-skill --dry-run
python3 setup_flowstate.py --clients claude-desktop,claude-code,cursor --schedule-backups --no-offsite --install-skill --verify

• Claude DesktopSetup above writes its entry for you. By hand, the entry in claude_desktop_config.json looks like this, with your own paths:

  {
    "mcpServers": {
      "flowstate": {
        "command": "/path/to/FlowState/.venv/bin/python",
        "args": [
          "-m",
          "flowstate"
        ],
        "env": {
          "FLOWSTATE_DB": "/path/to/flowstate.db",
          "PYTHONPATH": "/path/to/FlowState/mcp-server"
        }
      }
    }
  }

• Claude CodeSetup above adds it, or use claude mcp add as shown in the Claude Code memory guide. Source
• CursorCursor reads MCP servers from ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project), in the same format as the Claude Desktop entry above. Source On another computer, use your Relay address with a key in an environment variable:

  {
    "mcpServers": {
      "flowstate": {
        "url": "https://<handle>.fsmcp.com/mcp",
        "headers": { "Authorization": "Bearer ${env:FLOWSTATE_CURSOR_TOKEN}" }
      }
    }
  }

  Make the key with flowstate remote add-client --name "Cursor" on the computer where FlowState runs.
• claude.ai and the Claude appsWith Relay or Hosted, your address is https://<handle>.fsmcp.com/mcp. On Pro and Max: Customize, then Connectors, then Add, then Add custom connector. Enter a name and the address, then finish the sign-in and approve on the FlowState page. On Team and Enterprise an owner adds it first. Free plans can add one custom connector. Source
• ChatGPTOpenAI's help page says custom MCP servers are for ChatGPT Business and Enterprise/Edu on the web, with workspace permissions; Pro users can connect them with read and fetch only; phone apps cannot add them. Where allowed: Plugins, then the plus button, then Add custom MCP server, and enter your https address with sign-in. Source Because ChatGPT's access depends on your plan and workspace, we cannot promise every ChatGPT account can write to FlowState.
• GrokGo to grok.com/connectors, choose New Connector, then Custom, enter your https address and complete the sign-in. Source Then approve on the FlowState page.

What gets shared

Every connected app uses the same FlowState tools, so they all see the same records. Source

• ProjectsAn AI starting on a project gets a short brief of where things stand: open to-dos, problems, recent changes and hard rules.
• Decisions, with the whyDecisions, changes, problems and learnings are logged as the AI works, each marked with the app that wrote it, so you can see who did what.
• To-dosOne list any app can read, pick up and close. An agent can claim a to-do so two agents do not do the same work.
• HandoffsA note of where one app stopped and what comes next, for the next app to pick up.
• Review queueSuggestions and automatic session summaries wait for you. You approve or reject them by tapping a card, on the review page or your phone, or in a terminal. An AI never approves its own suggestions.

Questions

• Can Claude and ChatGPT share the same memory?Not on their own. Each app keeps its own memory in its own account. If both connect to the same MCP memory server, they read and write the same notes there. ChatGPT reaches a server only by a public https address, so the server has to be reachable that way.
• What is an MCP memory server?A program that AI apps connect to over the Model Context Protocol. It gives them tools to save and look up notes, such as decisions and to-dos, in a store that lives outside any one app or chat.
• Do I need Relay or Hosted?Only for apps that cannot start a program on your computer. Claude Desktop, Claude Code and Cursor can run FlowState locally for free. claude.ai, the Claude phone apps, ChatGPT and Grok in the browser connect from their own servers, so they need a public https address, which Relay or Hosted gives you.
• Does my data leave my computer?With local FlowState, no: your memory is one file on your computer. With Relay, it still stays on your computer and the relay passes encrypted traffic to it. With Hosted, your memory is kept on our server instead.
• Can an AI approve its own suggestions?No. Suggestions and automatic session summaries wait in a review queue. You approve or reject them yourself, by tapping a card, on the review page, on your phone, or in a terminal.
• Is FlowState the only memory server?No. Other MCP memory servers exist, with different storage and features. Any of them can be connected the same way. FlowState is free and open source (MIT).