Kontxt MCP Server

Name: Kontxt MCP Server
Author: ReyNeill

@ReyNeill

代码仓库

关于 Kontxt MCP Server

Codebase indexing MCP server

基本信息

分类

其他

运行时

python

传输方式

stdio

发布者

ReyNeill

配置

使用下面的配置,将此服务器添加到你的 MCP 客户端。

{
  "mcpServers": {
    "Kontxt": {
      "command": "python",
      "args": [
        "-m",
        "venv",
        "venv"
      ]
    }
  }
}

工具

未检测到工具

工具是从 README 中自动提取的。维护者可以在 ## Tools 标题下列出工具,即可填充这部分内容。

概览

What is Kontxt MCP Server?

Kontxt MCP Server is a Model Context Protocol server that indexes a user-specified local code repository and provides the get_codebase_context tool for AI clients like Cursor or Claude Desktop. It uses Google Gemini (default: gemini-2.5-flash-preview-04-17) internally to analyze the codebase and generate contextual responses based on the user’s query.

How to use Kontxt MCP Server?

Clone the repository, create a Python virtual environment, install dependencies from requirements.txt, and ensure the tree command is available on your system. Configure a Gemini API key in a .env file or via the --gemini-api-key argument. Run the server with python kontxt_server.py --repo-path /path/to/your/codebase. It starts in SSE mode on http://127.0.0.1:8080/sse by default; clients connect to this endpoint.

Key features of Kontxt MCP Server

Connects to a local code repository for analysis.
Provides the get_codebase_context tool for AI clients.
Uses Gemini 2.0 Flash (or custom model) with 1M input window.
Supports SSE (recommended) and stdio transport protocols.
Tracks token usage and allows configurable token limits (500k, 800k, 1M).
Accepts user-attached files and additional context from queries.

Use cases of Kontxt MCP Server

Ask an AI client “What’s this codebase about?” using the MCP tool.
“How does the authentication system work?” – context-aware explanation.
“Explain the data flow in the application” – targeted analysis.
Compare implementations across files by referencing them in queries.

FAQ from Kontxt MCP Server

What transport protocols does Kontxt MCP Server support?

It supports SSE (Server‑Sent Events, recommended) and stdio transport. Use --transport sse or --transport stdio.

What are the runtime requirements?

Python 3, the tree command, and a Google Gemini API key. The server uses the google/gemma-7b tokenizer for token estimation (gated on Hugging Face; free access requires accepting terms).

How do I configure CORS?

By default only loopback origins are allowed. Use --cors-origins with a comma‑separated list of origins, or set the KONTXT_CORS_ORIGINS environment variable. Credentials are off unless --cors-credentials is set.

How do I stop the server?

Press Ctrl+C in the terminal. The server attempts a graceful shutdown with a 3‑second timeout.

Can I change the Gemini model or token limit?

Yes. Use --gemini-model NAME to specify a different model (default: models/gemini-2.5-flash-preview-04-17) and --token-threshold NUM to set the token limit to 500000, 800000 (default), or 1000000.