Engram Is in the ChatGPT Plugin Directory — Here's How to Share Memory Across Platforms

Engram is now available in the ChatGPT Plugin Directory. You can connect it in under 30 seconds — no API key, no CLI, no install. Just OAuth and go.

But the real story isn't that Engram works with ChatGPT. It's that Engram gives you one memory across all your AI tools. Research something in ChatGPT, switch to Claude Code to build it, and the context carries over. That's the unlock.

Setup: 30 seconds, no API key

Connecting Engram to ChatGPT is the simplest integration we support:

1. Open ChatGPT → Settings → Plugins
2. Search for Engram or add a custom MCP connector
3. URL: https://mcp.getengram.app/mcp
4. Sign in with OAuth when prompted

That's it. ChatGPT now has tools to search your memory (search), store conversations (append_messages), and manage conversation threads (create_conversation, get_conversation). You can tell it “remember this” or ask “what did we discuss about X last week?”

The real unlock: cross-platform memory

Most people start using Engram with one tool. But the value compounds when you connect multiple tools to the same memory. Because Engram is an MCP server — not a plugin tied to one platform — any tool that speaks MCP can read from and write to the same store.

Today, Engram works with:

• ChatGPT— via the Plugin Directory (OAuth)
• Claude Code — via ~/.claude.json MCP config
• Claude Desktop— via MCP server config
• Cursor— via MCP settings
• Windsurf, Zed, Continue— any MCP-compatible client

All of them hit the same Engram server, the same account, the same conversations. When ChatGPT stores something, Claude Code can find it. When Cursor saves a debugging session, ChatGPT can recall it. There's no import/export step, no sync, no manual transfer. It's one memory.

Example: research in ChatGPT, execute in Claude Code

Here's a workflow that becomes natural with shared memory:

Step 1: Research in ChatGPT.You're on your phone or in a browser. You ask ChatGPT to research authentication patterns for your new API — compare JWT vs session tokens, evaluate OAuth libraries, look at how similar projects handle refresh tokens. ChatGPT does the research, you discuss tradeoffs, and you arrive at a decision: “Use jose for JWT validation, short-lived access tokens with rotating refresh tokens, store refresh tokens in D1.”

You tell ChatGPT: “Save this conversation to Engram.” It calls create_conversation and append_messages, storing the full research session with all the reasoning.

Step 2: Build in Claude Code.You sit down at your terminal. You open Claude Code in your project directory. Claude Code has Engram configured as an MCP server. You say: “I did some research on auth for this API — check Engram.”

Claude Code calls search with "authentication JWT refresh tokens". It finds the ChatGPT conversation from earlier — the full discussion, the options considered, the decision, the reasoning. Now Claude Code implements the auth layer with full context about why you chose this approach, not just what to build.

No copy-pasting. No “I decided earlier to use JWT because...” re-explanation. The context just flows from one platform to another through shared memory.

More cross-platform patterns

Mobile research, desktop execution

Use ChatGPT on your phone during your commute to research and plan. Store the conversation. When you get to your desk, your coding agent has the full context.

Design in one tool, implement in another

Discuss API design with Claude Desktop. Store the conversation. Open Cursor to implement it. Cursor's agent searches Engram and finds the design discussion with all the endpoint specs and edge cases you already thought through.

Debug across tools

Hit an error in Cursor. The investigation gets stored in Engram. Later, you ask ChatGPT about the same system and it can find the debugging session — even though it happened in a completely different tool.

Team handoffs

On a Team plan, one person's ChatGPT research is searchable by another person's Claude Code. The memory isn't per-tool or per-person — it's per-organization.

How it works under the hood

When you store a conversation, Engram:

1. Stores every message verbatim— no summarization, no lossy compression
2. Chunks the conversation into segments and generates vector embeddings using Cloudflare Workers AI
3. Indexes the chunks in Cloudflare Vectorize for semantic search and in FTS5 for keyword search
4. At search time, combines both signals with Reciprocal Rank Fusion (RRF) plus recency and title-match boosts

This means you can search by meaning, not just keywords. Searching for “that database migration issue” finds the relevant conversation even if you never used those exact words. And it works across every tool connected to your account.

Getting started

If you're already using ChatGPT, you can connect Engram in under a minute:

1. Sign up free at getengram.app
2. Add Engram in ChatGPT Settings → Plugins
3. Optionally, add it to Claude Code, Cursor, or any other MCP client

The free tier gives you 10,000 messages of memory that never expires — enough to try cross-platform memory on a real project and see if it changes how you work. It probably will.