Kimi Setup (Kimi Code CLI & Claude Code)

Overview

Kimi is Moonshot AI's model family; the newest flagship, Kimi K3, launched on July 16, 2026. There are two working routes to pair Kimi with Tempreon: the native Kimi Code CLI, or running Kimi as the model inside Claude Code. Both give you the same result — your AI loads your Core Imprint, searches your Knowledge Vault, and remembers what it learns, across every tool you connect.

Which route to pick: if you already use Claude Code, route B slots Kimi into your existing setup with two environment variables. If you want Moonshot's own harness, route A is a two-command setup.

Quick path: kimi mcp add --transport http tempreon <URL> → kimi mcp auth tempreon → approve in the browser → done.

Prerequisites

• A Moonshot account with Kimi Code access (route A) or a Moonshot API key (route B)
• A Tempreon account with at least one available Bridge slot
• A working default browser for the authorization step

Step-by-Step Setup

Route A: Kimi Code CLI (native)

Add Tempreon from your terminal:

kimi mcp add --transport http tempreon https://api.tempreon.com/functions/v1/tempreon-mcp/mcp

The URL is also shown in your Bridges dashboard. Then authorize:

kimi mcp auth tempreon

A browser tab opens with the Tempreon consent screen — sign in if prompted and approve. Inside the TUI, /mcp shows server status and /mcp-config manages the configuration.

One caution: the underlying config file's location has moved between Kimi CLI versions, so don't rely on editing it by hand — the CLI commands above are the reliable path, and kimi mcp list confirms what's actually registered.

Route B: Kimi inside Claude Code

Moonshot runs an Anthropic-compatible endpoint, so Claude Code can drive Kimi directly. Point Claude Code at it:

export ANTHROPIC_BASE_URL=https://api.moonshot.ai/anthropic
export ANTHROPIC_AUTH_TOKEN=your-moonshot-api-key

and use the kimi-for-coding model. Your existing Claude Code MCP setup — including Tempreon — keeps working unchanged, because MCP lives in the harness, not the model. If you haven't connected Tempreon in Claude Code yet:

claude mcp add --transport http tempreon https://api.tempreon.com/functions/v1/tempreon-mcp/mcp

then run /mcp inside Claude Code to complete the OAuth login. The full walkthrough is in the Claude Code guide.

Route C: kimi.com

As of July 2026, the consumer web app at kimi.com has no custom connector path — you can't add Tempreon there. Use route A or B; anything your AI learns through them is waiting in your Tempreon account regardless of which surface you work from.

Troubleshooting

Connection fails immediately. Check the URL ends in /mcp — no trailing slash. The full URL is https://api.tempreon.com/functions/v1/tempreon-mcp/mcp.

Hand-edited config isn't being picked up. The config file's location varies across Kimi CLI versions. Re-add the server with kimi mcp add and verify with kimi mcp list instead of editing files directly.

Auth loop or stale tokens (route A). Run kimi mcp reset-auth for Tempreon, then kimi mcp auth tempreon again for a fresh browser flow.

Kimi responds but Tempreon tools are missing (route B). The env vars only swap the model — the Bridge is configured separately. Confirm Tempreon shows as connected under /mcp in Claude Code, and restart the session after changing environment variables.

Also connect: GLM · OpenCode · Qwen Code

Try saying...

• "Start my session." — triggers session_start to load your Core Imprint
• "Remember: we deploy Fridays only after the staging soak passes." — stores a working rule your AI keeps across models
• "Search my knowledge base for notes on the billing refactor." — exercises retrieval end-to-end