kubbi-ai — Claude Code MCP server: what it does, how to install
kubbi-ai is a Claude Code MCP server for passing agent payloads through temporary encrypted storage using the claim-check pattern, with burn-after-read deletion.
no license 157 installs updated 2026-05-17
claude mcp add --transport http kubbi-ai https://kubbi-ai--admin-dlwa.run.toolsWhat kubbi-ai does
The server moves bulky or sensitive payloads out of the direct prompt channel by swapping them for compact claim tokens. One agent pushes content into encrypted storage, routes the resulting URL through an MCP result, webhook, or queue, and another agent fetches decrypted data via that token. Short expiry windows, resting encryption, and burn-after-read destruction work by default without extra toggles.
Inside, the server exposes six operations. Senders use kubbi_send for single payloads or kubbi_send_files to bundle at most 5 files up to 5 MB total, encoding binary items as base64 strings. Receivers inspect metadata and manifests safely through kubbi_inspect without consuming retrieval quotas, or pull final content via kubbi_claim. Senders can also track IDs with kubbi_get and wipe records instantly with kubbi_delete.
Publishing or deleting payloads strictly requires KUBBI_API_KEY. Receiving agents need zero API credentials to inspect metadata or decrypt the underlying payload.
What it helps with
- offload agent payloads into compact claim tokens instead of swelling message limits
- package up to 5 files totaling up to 5 MB into a single encrypted bundle
- wipe sensitive payload data automatically once the recipient finishes reading
- inspect file manifests and expiration timers without burning the retrieval limit
Example prompts for the agent
- Create an ephemeral payload with kubbi_send and return the claim URL
- Check the token status using kubbi_inspect without consuming the payload
- Retrieve and decrypt the package content using kubbi_claim
How to install kubbi-ai
Platform CLI: remote server
claude mcp add --transport http kubbi-ai https://kubbi-ai--admin-dlwa.run.toolsPlatform CLI: remote server
codex mcp add kubbi-ai --url https://kubbi-ai--admin-dlwa.run.toolsConfig file
{
"mcpServers": {
"kubbi-ai": {
"url": "https://kubbi-ai--admin-dlwa.run.tools"
}
}
}Config file
{
"servers": {
"kubbi-ai": {
"type": "http",
"url": "https://kubbi-ai--admin-dlwa.run.tools"
}
}
}Platform CLI: remote server
gemini mcp add -t http kubbi-ai https://kubbi-ai--admin-dlwa.run.toolsStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about kubbi-ai
How to install kubbi-ai in Claude Code?
Run claude mcp add --transport http kubbi-ai https://kubbi-ai--admin-dlwa.run.tools in your terminal.
What does kubbi-ai do?
It provides an encrypted temporary storage layer where agents publish payloads, share lightweight claim URLs, and retrieve content with automatic burn-after-read expiration.
Does the consuming agent need an API key?
No, retrieval tools such as kubbi_inspect and kubbi_claim require zero credentials, while sending and deletion tools demand KUBBI_API_KEY.