claude-context (github.com/zilliztech/claude-context), open-sourced by vector database company Zilliz, is becoming one of the hottest MCP (Model Context Protocol) projects on GitHub. It has reached 10,200 stars with approximately 3,700 new stars this week, 194 commits, and 39 open PRs.
The Problem It Solves
While Claude Code and other coding agents have increasingly large context windows, stuffing entire codebases of hundreds of thousands of lines is neither economical nor efficient. claude-context’s approach is vector retrieval instead of context stuffing:
- Index the codebase into a vector database (supports Milvus/Zilliz Cloud)
- Provide code search capability to Claude Code via MCP protocol
- Agent retrieves relevant code snippets on demand rather than passively receiving all context
This essentially adds an “external memory layer” for coding agents.
Core Capabilities
The project uses a multi-package architecture supporting:
- Semantic Search: Natural language queries to locate relevant code files and functions
- Cross-file Relations: Tracking dependencies and call chains for complete code context
- Multi-model Support: Compatible with Gemini and other LLMs, not just Claude
- Deduplication: Recent PR (#333) adds search result deduplication
- Evaluation Framework: Built-in evaluation directory with case studies and benchmarks
Latest version is v0.1.3 with Node.js 24 support.
Quick Start
git clone https://github.com/zilliztech/claude-context.git
cd claude-context
npm install
npx claude-context serve --milvus-url http://localhost:19530
Requires local or cloud deployment of Milvus vector database. Docker Compose one-click startup is provided in the documentation.