Home › Connect your AI client

Connect Kagura Memory to your AI client

One server URL connects Kagura Memory Cloud to Claude, ChatGPT, Codex, Cursor and other MCP clients. This page covers what you need, how to connect each client, prompts to try, what the tools do, and what to do when something goes wrong.

What it does

Kagura Memory Cloud is persistent memory for AI assistants. When you ask, your assistant saves a note — a decision and its reason, a fact, a to-do — and finds it again later: in a new conversation, or from a different AI tool connected to the same account.

It works over the Model Context Protocol (MCP). Your AI client connects to one server URL, you sign in with your Kagura account, and the assistant gets a set of memory tools. Memories are stored in your own workspace, grouped into contexts (for example one per project or topic), and stay private to you unless you share a context with your team.

Before you connect

Request an invite See pricing

Server URL and sign-in

Every client uses the same server URL:

https://memory.kagura-ai.com/mcp

The transport is Streamable HTTP. Claude, ChatGPT, Codex and Cursor sign in with OAuth, so there is no key to paste. For other clients, see Other MCP clients below; some of them need an API key instead.

Signing in

The first time a client connects, it opens a Kagura sign-in page in your browser. Sign in with Google, with GitHub, or with your email address and password — password sign-in is available once you have set a password under Profile in the Kagura web app.

The approval screen

After you sign in, Kagura shows an authorization screen. It names the app asking for access, lists what the app will be allowed to do with your memories (read them, write new ones, delete them), shows the account you are signing in as, and shows the address you will be sent to after you approve.

Check that the address belongs to the AI client you are connecting, then choose Authorize. Cancel stops the connection and shares nothing. If the app runs on your own computer, such as Claude Code or the Codex CLI, the screen says so: approve only if you just started the sign-in yourself.

Connect from your client

Pick your client below. Menu names change from time to time; if yours look different, the client's own help page linked in each section has the current steps.

Claude (claude.ai and Claude Desktop)

  1. Open Customize → Connectors.
  2. Select + and then Add custom connector. Enter the server URL https://memory.kagura-ai.com/mcp (and a name, such as Kagura Memory, if asked), then select Add.
  3. Select Connect. Sign in to Kagura in the window that opens and approve access.
  4. In a conversation, open the + menu → Connectors and make sure Kagura Memory is turned on.

Claude Desktop uses the same connectors: remote connectors are configured through your Claude account, so one added on claude.ai is also available in the desktop app. On the Free plan you can add one custom connector. On Team and Enterprise plans, an Owner first adds it under Organization settings → Connectors (Add → Custom → Web); each member then selects Connect under Customize → Connectors.

Claude Help Center: custom connectors

Claude Code

  1. Add the server from your terminal: 
    claude mcp add --transport http kagura-memory https://memory.kagura-ai.com/mcp
  2. In Claude Code, run /mcp, choose kagura-memory and select Authenticate. Sign in and approve access in your browser.
  3. Run /mcp again to check that kagura-memory is connected.

Add --scope user to the command to use the server in all your projects; without it, the server is added for the current project only.

Claude Code docs: MCP

ChatGPT

  1. On chatgpt.com, turn on developer mode: Settings → Security and login → Developer mode.
  2. Go to Settings → Plugins (called Apps or Connectors in earlier versions) and select +. Enter a name and description, enter the MCP server URL https://memory.kagura-ai.com/mcp under Connection, and create the connection.
  3. When ChatGPT asks, sign in to Kagura and approve access.
  4. In a chat, open the + menu, choose Developer mode, and select Kagura Memory.

Developer mode is available on Plus, Pro, Business, Enterprise and Education plans on the web. ChatGPT asks you to confirm each tool call that changes data; check what will be sent before you approve.

OpenAI docs: ChatGPT developer mode

Codex

  1. Add the server from your terminal: 
    codex mcp add kagura-memory --url https://memory.kagura-ai.com/mcp
  2. Sign in and approve access in your browser: 
    codex mcp login kagura-memory
  3. Run codex mcp list to check that kagura-memory is there, then start Codex.

The Codex CLI, the Codex IDE extension and the ChatGPT desktop app share this configuration (~/.codex/config.toml). In the IDE extension or the desktop app you can also add the server under the settings (gear) menu → MCP servers → Add server.

OpenAI docs: MCP in Codex

Cursor

  1. Add the server to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project): 
    {
  "mcpServers": {
    "kagura-memory": {
      "url": "https://memory.kagura-ai.com/mcp"
    }
  }
}
  2. Open Customize in Cursor's sidebar and check that kagura-memory is turned on.
  3. When Cursor asks you to sign in, complete the Kagura sign-in in your browser and approve access.

Cursor docs: MCP

Other MCP clients

Any client that supports remote MCP servers over Streamable HTTP can connect.

  1. Add a remote MCP server with the URL https://memory.kagura-ai.com/mcp.
  2. Sign in to Kagura when prompted and approve access.

Kagura accepts automatic sign-in registration from Claude, ChatGPT, Codex and Cursor, and from Hermes Agent and OpenClaw. If your client's sign-in is refused, create an API key in the Kagura web app (API keys) and use the connection details shown there instead. Keep the key secret, like a password.

Example prompts

Once connected, try these. You don't need to name the tools — the assistant chooses them. If it doesn't use Kagura Memory, add "using Kagura Memory" to the request.

Remember that we chose PostgreSQL for the billing service, because invoices and payments must be updated in one transaction.

Saves the decision together with its reason.

What did we decide about the billing database, and why?

Searches your memories and answers from what was saved — also in a new conversation or another connected client.

List what you remember about this project.

Finds the project's context and summarizes the memories in it.

Remember this as a standing rule for this project, so you always load it first: we don't deploy on Fridays.

Saves a pinned memory, which the assistant loads at the start of a session instead of waiting for a search to find it.

Forget the note about the old staging server.

Finds the note and deletes it. Your client will usually ask you to confirm first.

What the tools do

Your assistant calls these tools on your behalf. Each tool tells the client whether it only reads, changes, or deletes data, and clients use that to decide when to ask you to confirm. What a tool may do is also limited by your role in the workspace.

Save Changes data

Save a new memory, or correct or update one that is already saved.

rememberupdate_memory

Search Changes data (ranking)

Search memories by meaning and keywords. Each search also updates how memories are ranked in later searches, so recall is marked as changing data, and some clients ask you to confirm before running it.

recall

Read Reads

Read one memory in full, follow links to related memories, load the pinned set, and list your contexts and tags.

referenceexploreload_pinnedlist_contextsget_context_infolist_tags

Organize and link Changes data

Create and edit contexts, merge one context into another, and link related memories to each other (for example, mark that a new decision replaces an old one).

create_contextupdate_contextmerge_contextscreate_edgeupdate_edge

Recall by time and place Reads

List what is coming up (reminders and dated notes), or find memories saved with a location near a given place.

recall_upcomingrecall_nearby

Delete and forget Deletes data

Delete a memory, a link between memories, or a whole context with all its memories. A forgotten memory is removed permanently after a retention period.

forgetdelete_edgedelete_context

The server also has tools for files, agent state, analyses, secrets and workspace administration; many of them need a particular plan or role.

Long lists and long memories come back in pages, so a single response never gets too large for the client. The assistant can ask for the next page.

Troubleshooting

Sign-in keeps looping, or the client says the session expired

Disconnect Kagura Memory in your client and connect it again, then sign in and approve access. In Claude Code, run /mcp, choose kagura-memory and select Authenticate (or Clear authentication first). In Codex, run codex mcp login kagura-memory.

A tool fails with "Too many requests"

You've reached a short-term rate limit. Wait a minute and try again.

If the message says a daily limit was reached — such as memories created per day or tool calls per day — the limit resets at 00:00 UTC. Each plan's limits are listed under Pricing.

A tool is refused for permission or scope

"insufficient_scope" means the access you approved doesn't cover that tool (for example, read-only access and a tool that saves). Reconnect and approve the access the client asks for.

"permission_denied" means your role in the workspace or context doesn't allow the action. Ask the workspace owner.

Results look cut off

Long lists and long memories are returned in pages. Ask the assistant to get the next page or to read that memory in full, or narrow the request — name a context, a tag or a date range.

The assistant doesn't use Kagura Memory

Check that the connector is turned on for this conversation (in Claude: + → Connectors; in ChatGPT: + → Developer mode), then mention it by name: "Using Kagura Memory, …".

How do I disconnect?

Remove the server in your client. Claude: Customize → Connectors, then Remove (or the ⋯ menu). Claude Code: claude mcp remove kagura-memory. Codex: codex mcp remove kagura-memory. Cursor: delete the entry from mcp.json. ChatGPT: open the plugin under Settings → Plugins and choose Uninstall plugin.

Disconnecting doesn't delete your memories; they stay in your workspace.

How do I delete memories or my account?

To delete a memory, ask your assistant to forget it, or delete it in the Kagura web app. A forgotten memory is removed permanently after a retention period.

To delete your account and all its data, open Profile → Delete account in the Kagura web app. Deletion starts after a 7-day cooling-off period, during which you can cancel. If you can't sign in, contact us using the address below.

Privacy and support

Memories are stored in your Kagura workspace and are private to you unless you share a context with your team.