kimi-code-mcp
An MCP server that hands large-repo analysis and independent diff review to Kimi Code so Claude spends fewer tokens reading files.
- 63 ★ on the repository
- Last commit
Extended access
Why this level
- Repository contents and diffs are sent to the Kimi service and the API key lives in the environment.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/kimi-code-mcpDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Repository contents and diffs are sent to the Kimi service and the API key lives in the environment.
This is third-party code. Review the repository files before installing.
What it does
`kimi-code-mcp` connects Kimi Code to Claude Code as a set of MCP tools. Codebase analysis (`kimi_analyze`) reads the repository from disk through the Kimi CLI and returns a summary, and `kimi_resume` continues the same session with follow-up questions. A quick question with no context (`kimi_query`) and independent review of a diff or claim (`kimi_verify`) go straight to the Kimi Code API with a key. There is also `kimi_status`, which shows which backends are available, and local tools for sessions and cache. There are eight tools in total. The default model is `k3`, changed with the `model` argument or the `KIMICODE_MODEL` variable. The author cites about 90% Claude-side token savings on analysis tasks and calls the numbers illustrative.
Who it is for. Claude Code users with a Kimi Code subscription who want to offload reading of large repositories and get a second opinion from another model.
Good fit when
- A monorepo of hundreds of files must be analyzed before a refactor.
- Before a commit an independent review of the diff by another model is needed.
- A short programming question goes to a second model without project context.
- After analysis you want to continue the same session with follow-ups.
Not a fit when
- Code must not be sent to the external Moonshot service.
- There is no Kimi Code subscription and no plan to pay for one.
- The task is small and the overhead of a second agent outweighs the benefit.
Example request
Analyze the structure of this monorepo through kimi_analyze, then check my diff through kimi_verify and show the problems found.MCP
- Transport
- stdio
- Authentication
- key is optional
Environment variables
KIMICODE_API_KEYsecret- Kimi Code API key for the `kimi_query` and `kimi_verify` tools
KIMICODE_MODEL- Default model instead of `k3`
Limitations
A Kimi Code subscription is required: `kimi_query` and `kimi_verify` need an API key while `kimi_analyze` and `kimi_resume` also need the installed Kimi CLI signed in with `kimi login`. Repository contents and diffs go to the Kimi service. The README says that on 2 November 2026 npm holds version 0.3.0 and version 0.5.0 from git is unpublished, so the newer behavior comes from a source install. Token savings are the author's estimate and depend on the task. It relies on the `~/.local/bin/kimi` path.
How to disable. Run `claude mcp remove kimi-code` or delete the entry from `.mcp.json`. Remove the package with `npm uninstall -g kimi-mcp-server` and revoke the API key in the Kimi Code console.
FAQ
How do I connect it to Claude Code?
Run `claude mcp add --scope user --transport stdio kimi-code -- npx -y kimi-mcp-server` or add the server to the project `.mcp.json`.
Which tools need the CLI and which need a key?
`kimi_analyze` and `kimi_resume` work through the CLI. `kimi_query` and `kimi_verify` call the API and need only a key.
Where do I get an API key?
In the Kimi Code console. The server reads `KIMICODE_API_KEY`, then `KIMI_CODE_API_KEY` and `KIMI_API_KEY`.
How do I change the model?
Pass the `model` argument or set `KIMICODE_MODEL`; the argument wins.
How do I find out why a call failed?
Run `kimi_status`: it reports the API key state and CLI login state separately.
Discussion
No questions or reviews yet.
Related
MCP serversKimi Code is Moonshot AI's terminal coding agent: it reads the project, edits files and runs commands. Installs with one script, no Node.js needed.
Meta_Kim guides an agent from idea to verified result: it gathers context, plans, assigns tasks and writes lessons back.
A terminal coding agent forked from Kimi CLI, with lean prompts, Python scripting and support for dozens of API providers.
An MCP server that preserves Claude Code context between sessions: decisions, checkpoints and a file cache kept in local SQLite.
An MCP bridge: a single run_python tool runs code in a container and calls your MCP servers
Plugin and MCP server that asks GPT, Gemini, Grok and local models to review a plan or hunt for bugs before code ships.
README in short
The README covers a savings table, quick start, two ways to reach Kimi, key and login setup, eight tools and advice for agents on when to call which.