← All docs

// Docs

Klyr in claude.ai.

Klyr lives primarily in Claude Code on your machine. But your clone's memory is hosted, and claude.ai can reach it directly: add Klyr as a custom connector and the same memory — dynamic context, durable conclusions, your knowledge archive — is available from any browser, your phone, Claude Desktop, and claude.ai cloud code sessions. Nothing to install.

Add the connector

1. In claude.ai, open Settings → Connectors.
2. Choose Add custom connector.
3. Enter the server URL: https://api.getklyr.ai/mcp
4. Click Connect. Claude sends you to the Klyr sign-in page.
5. Enter the email you purchased Klyr with. We email you a one-time sign-in link — open it, and the connection finishes on its own. Nothing to copy or paste.

That is the same sign-in used by the plugin: your purchase email is your identity, and the link expires in 15 minutes. Once connected, the connector stays signed in; you will only re-approve occasionally.

On mobile, connectors added on the web are already there — one account, one connection.

Wake your clone

A claude.ai conversation does not start with your clone's identity loaded — that injection normally happens on your machine, at session start. The connector ships a wake prompt that does the equivalent in one gesture:

1. In the composer, open the + (attachments) menu and pick the Klyr connector's prompts.
2. Choose wake.

It loads your persona kernel (SOUL, USER, MEMORY) and the current memory context into the conversation, so the assistant speaks as your clone from the first message. Use it at the start of any conversation where you want the full persona, not just the memory tools.

If the prompts menu is not available on your surface, the fallback is a claude.ai Project: paste your kernel files into the Project's instructions once, and every conversation in that Project starts with them.

What works, and what stays in the terminal

Through the connector you get the full memory tool surface: recall (reasoning, context, search, llmbrain_search), capture (conclude, profile), and kernel sync. Ask "what do you know about X" or "remember this" from your phone and it lands in the same memory your terminal clone uses.

What the connector does not do — these live in the Claude Code plugin, which has a runtime on your machine:

• Automatic context injection per message. On claude.ai, memory is fetched when the conversation needs it (or when you use wake), not on every turn.
• Conversation capture. claude.ai conversations are not swept into memory. conclude still writes anything worth keeping, on demand.
• Skills, commands, and hooks (/klyr:*, the action journal, the statusline). The connector is a memory surface, not a runtime.

Cloud code sessions

Sessions on claude.ai/code run in Anthropic-managed sandboxes. Connector traffic is routed through Anthropic's servers, so enabling the Klyr connector on a session gives it your clone's memory with zero sandbox configuration — no keys to paste, no domains to allow.

Troubleshooting

• "Connection failed" when adding the connector. Check the URL is exactly https://api.getklyr.ai/mcp, then retry. If it persists, see troubleshooting or ask on the help page.
• No sign-in email. Check spam, and make sure you entered the email you purchased with — for privacy, the page looks identical whether or not the address is known.
• Tools return "no memory yet". Normal for a fresh account or a peer with no captured memory; it fills as you use Klyr.
• wake says there is no synced kernel. Sync it once from a device where the Klyr plugin is installed, then run wake again.