How to share memory between Claude, ChatGPT and Cursor

• ChatGPT: saved memories and references to past chats, inside your ChatGPT account.
• Claude (web, desktop and mobile): its own context lives inside your Claude account.
• Claude Code: CLAUDE.md files and an auto memory on your machine, per project.
• Cursor: project rules (.cursor/rules), user rules and AGENTS.md. Its documentation puts it this way: models keep no memory between completions, and rules are the context that gets repeated.

None of these memories can be read from the other tools.

Pasting the same text into each tool’s instructions works for what does not change. But new decisions come up in the middle of work, in one tool, and nobody copies them to the others.

Illustrative example: in a ChatGPT conversation you decide prices are shown in pesos with VAT included. The next day, the Cursor agent writes the pricing page without VAT, because that decision never reached its rules.

MCP (Model Context Protocol) is the standard these tools use to connect to external services, and all four support remote MCP servers. If they all connect to the same memory server, what one stores the others find.

NEXUS is that server: https://nexus.eblas.link/mcp. Whatever any of your agents stores (a decision, a rule, a mistake and how it was fixed) stays in your account and the others recall it before acting. Each account is isolated from the rest.

First create a free account at https://nexus.eblas.link/en/signup. Then connect each tool to https://nexus.eblas.link/mcp:

• Claude: Customize → Connectors → add a custom connector with that address and authorize.
• ChatGPT: Settings → Apps → Advanced settings → Developer mode. Create an app named “Nexus AGI” with that address and OAuth authentication.
• Claude Code: claude mcp add --transport http nexus-agi https://nexus.eblas.link/mcp, then /mcp to authorize.
• Cursor: in ~/.cursor/mcp.json (or the project’s .cursor/mcp.json) add the server with an agent credential you generate in the NEXUS console, under Credentials:

{
  "mcpServers": {
    "nexus-agi": {
      "url": "https://nexus.eblas.link/mcp",
      "headers": { "Authorization": "Bearer nxa_..." }
    }
  }
}

In Claude and ChatGPT, also add the instructions from section 4 of the documentation (https://nexus.eblas.link/en/docs#s3): without them, the agent almost never checks memory on its own.

• Decisions, with their reason: “payment records are archived, not deleted, because accounting audits them”.
• Business rules a new agent cannot guess.
• Mistakes that already happened and how they were fixed.
• When something stops applying, say so: NEXUS marks the old memory as no longer valid instead of deleting it.

Do not store passwords, API keys or card data.

• Each connected tool counts as one agent. The Free plan allows 2 agents per month; to connect more, the Personal plan allows 3.
• NEXUS does not read your ChatGPT history or Claude’s memory: it only receives what your agents send it through its tools.
• A memory can go stale. Correct it when it happens.

No. Each keeps its memory in its own account and does not read the other’s. To share decisions between them, both have to connect to the same external memory, for example an MCP server.

Cursor works with rules: project rules in .cursor/rules, user rules and AGENTS.md. They are context repeated on every completion and are maintained by hand.

No. NEXUS only receives what your agent sends it through its tools: it does not read chat histories or other tools’ native memory.

The Free plan allows 2 agents per month, and each connected tool counts as one. The Personal plan allows 3.