kimi-code-mcp

Overview

kimi-code-mcp is a specialized MCP server that enables Claude Code to delegate bulk codebase analysis to Kimi K2.5, leveraging its 256K context window and cost efficiency. The server provides 7 tools including deep codebase analysis, quick querying, session management, and caching capabilities. It's designed to reduce Claude token consumption by 60-80% on analysis-heavy tasks while maintaining high code intelligence through Kimi's specialized coding model.

Try asking AI

After installing, here are 6 things you can ask your AI assistant:

• kimi_analyze

  Deep codebase analysis (architecture, audit, refactoring)

  Try:Analyze this codebase's architectureTry:Scan for security vulnerabilitiesTry:Map all dependencies of the auth module
• kimi_query

  Quick programming questions, no codebase context

  Try:How do I implement pagination in React?Try:Explain the singleton patternTry:Best practices for error handling
• kimi_list_sessions

  List existing Kimi sessions with metadata

  Try:Show my previous Kimi sessionsTry:List all available analysis sessions
• kimi_resume

  Resume a previous session (up to 256K token context)

  Try:Continue my last session about API designTry:Resume the codebase analysis from yesterday
• kimi_status

  Check CLI installation, version, and authentication status

  Try:Check my Kimi CLI statusTry:Verify Kimi authentication
• kimi_cache_status

  View session cache statistics and performance metrics

  Try:Show my Kimi cache statsTry:Check session performance metrics
• kimi_cache_invalidate

  Manually invalidate cached sessions (by dir or all)

  Try:Clear all cached Kimi sessionsTry:Invalidate cache for current directory

Comparable tools

Installation

curl -L code.kimi.com/install.sh | bash
kimi login

npm install -g kimi-mcp-server

{
  "mcpServers": {
    "kimi-code": {
      "command": "npx",
      "args": ["-y", "kimi-mcp-server"]
    }
  }
}

FAQ

Does this require a Kimi Code subscription?

Yes, you need an active Kimi Code plan (Moderato $19/mo or higher). The free 7-day trial is available for testing.

How does this save tokens compared to using Claude directly?

Kimi reads entire codebases (up to 256K tokens) at near-zero cost, while Claude only spends tokens on reasoning and precise code edits, reducing total token consumption by 60-80% for analysis tasks.

Compare kimi-code-mcp with