Kimi Code Usage

English · 中文

Kimi Code Usage: The Curated Toolchain

Manifesting your AI quota with aesthetic precision across CLI, MCP, and VS Code.

-100%25%20coverage-brightgreen)

🌟 Project Vision

In the era of "Vibecoding," transparency of resources is a prerequisite for flow. Kimi Code Usage is a meticulously crafted toolchain — three components, one soul.

⚠️ 重要：API Key 类型说明

本工具查询的是 Kimi Code 平台（Coding Plan 会员） 的用量数据。你需要的是 Kimi Code 控制台 创建的 API Key（格式 sk-kimi-xxx）， 不是 Kimi 开放平台（platform.kimi.com）的 API Key（格式 sk-xxx）。两种 Key 不互通。

获取方式：登录 Kimi Code 控制台 → 创建 API Key → 设置环境变量 KIMI_API_KEY。 详见 Kimi Code 文档。

Prerequisite: A Kimi Coding Plan API Key (sk-kimi-xxx), set as KIMI_API_KEY in your environment or .env file.

Monthly membership usage is optional. In interactive mode, if an installed Kimi WebBridge daemon is stopped, kimi-usage asks on every launch whether to start it. Declining affects only monthly credits. The browser extension must still be connected and Kimi must be logged in; weekly and 5-hour Kimi usage continue working when monthly usage is unavailable.

ChatGPT Plus usage reads the local Codex authentication file. After a 401 response, it refreshes the access token and retries once. If authentication still fails, run codex login.

⚡ CLI Reporter

A Rich-rendered panel in your terminal. Zero noise, pure signal.

Install & Run:

pip install kimi-code-usage
kimi-usage              # Aesthetic Rich panel
kimi-usage -i           # Interactive panel with optional WebBridge startup prompt
kimi-usage --json       # Machine-readable JSON
kimi-usage --plain      # Plain text output

Or run instantly without installing:

uvx kimi-code-usage

🔍 MCP Server

Exposes get_kimi_usage to any MCP-compatible AI Agent.

Compatible with Claude Code, Cursor, Windsurf, Hermes, and any MCP-enabled agent.

Add to your MCP config (e.g., ~/.claude/settings.json):

{
  "mcpServers": {
    "kimi-code-usage": {
      "command": "uvx",
      "args": ["--from", "kimi-code-usage", "kimi-mcp"],
      "env": {
        "KIMI_API_KEY": "YOUR_KEY"
      }
    }
  }
}

Note: The MCP server still needs --from because kimi-mcp is a separate command from the default kimi-code-usage entry point.

Then simply ask your AI: "Check my Kimi quota."

💎 VS Code Extension

A sleek status bar indicator with sensory color alerting.

Install: Search Kimi Code Usage in the VS Code Marketplace, or:

code --install-extension HainingYu.kimi-code-usage

Configure (Settings > Kimi Code Usage):

Usage: Status bar shows 🌕 [===] Wee:96% 5Ho:99%. Hover for details. Cmd+Shift+P → Kimi: Refresh.

📊 Real-World Analysis

How much does a ¥199/mo Coding Plan actually cover? Real data from 5 months of heavy vibe coding, with a cost comparison against standard API pricing.

Full data → docs/kimi-code-usage.md

1. Profile

2. Token Composition

A 91.9% cache hit rate is the signature of vibe coding — long sessions where each turn carries full conversation history, and only the latest exchange is fresh.

3. Actual Monthly Burn

Based on continuous weekly sampling (3 days = 30% of weekly quota):

4. Cost Comparison

Priced against Kimi K2.6 standard API rates (input ¥6.50 / cache ¥1.10 / output ¥27.00 per million tokens):

5. Bottom Line

The Coding Plan saves heavy users ≈ ¥2,221/month (~92%). Cache reads account for 92% of total tokens; the plan bundles both cache and input into the subscription fee. Under pay-as-you-go, cache alone makes up 63% of the bill.

⚠️ Disclaimer

This analysis reflects one user's actual usage from 2026-01 to 2026-05 via ccusage. Token ratios and cost estimates are for reference only — actual results vary by session pattern, model choice, caching behavior, and API pricing. Rates sourced from platform.kimi.com as of 2026-05-24.

🎨 About the Curator

Crafted with ❤️ by Haining Yu, an Art Curator and Vibecoder. This toolchain is part of a curated collection designed to bridge the gap between aesthetic curation and intuitive, AI-powered coding.