Claude Context
github.com/zilliztech/claude-context- Category
- Developer Tools
- Rank
- No. 52Tools index
- Type
- APP
- Builder
- zilliztech
- GitHub
- 12.5k stars
- Date
About
A code-search MCP server that makes your entire codebase available as context to Claude Code and other coding agents.
What it does
Claude Context indexes source files into chunks, converts them into embeddings, and stores them in a vector database. Coding assistants can then submit natural-language queries and receive relevant code with paths, line ranges, languages, and relevance scores.
Why it's ranked here
The design suits large repositories where repeatedly loading directories wastes context. Syntax-aware splitting, filtered retrieval, persistent indexing state, and automatic change synchronization make it more than a thin search wrapper. The required embedding and vector-database setup keeps it from being a lightweight default.
What's good
Indexing supports syntax-aware splitting with automatic fallback, custom extensions, and extra ignore patterns. Search results can be limited or filtered by extension. Background synchronization updates added, removed, and modified files, while global locking prevents multiple server processes from synchronizing simultaneously.
Tradeoffs
It requires Node.js 20 or newer, a Milvus vector database, and a configured embedding provider. The documented quick start needs separate Zilliz Cloud and OpenAI credentials. Default indexing excludes many common configuration, web, shell, and data formats unless users add extensions explicitly. MCP operations also expect absolute codebase paths.
How to use it well
Use it for repeated architecture discovery, refactoring, code review, and feature work across repositories too large to load wholesale. Index each repository, let background synchronization maintain it, then search with focused natural-language prompts and extension filters. It does not provide persistent agent memory across sessions; the documentation points to a separate tool for that need.
Technical notes+
The private pnpm monorepo in package.json builds separate core, MCP, VS Code, Chrome, and example surfaces. packages/mcp/src/index.ts exposes four stdio MCP tools: index_codebase, search_code, clear_index, and get_indexing_status. packages/core/src/context.ts defaults to AstCodeSplitter, supports custom extensions and ignore patterns, and derives distinct Milvus collection names from resolved codebase paths. packages/mcp/src/sync.ts runs background reindexing with an in-process guard plus a global filesystem lock. packages/mcp/src/snapshot.ts persists indexing state and migrates legacy snapshot formats. packages/mcp/src/config.ts supports OpenAI, VoyageAI, Gemini, Ollama, and OpenRouter embeddings.
Observed
- License
- MIT
- Primary language
- TypeScript, with JavaScript build configuration and asset scripts
- Runtime
- Node.js 20 or newer; pnpm 10 or newer for repository development
- Packaging
- Public npm packages for the core library and MCP server, plus a VS Code Marketplace extension
- Interfaces
- Core library, stdio MCP server, VS Code extension, and Chrome extension
- MCP tools
- Index codebase, search code, clear index, and report indexing status
- Embedding providers
- OpenAI, VoyageAI, Gemini, Ollama, and OpenRouter
- Documented clients
- Claude Code, Codex CLI, Gemini CLI, Qwen Code, Cursor, Void, Claude Desktop, Windsurf, VS Code, Cherry Studio, Cline, Augment, and Roo Code
Read from README.md, package.json, packages/vscode-extension/copy-assets.js, packages/chrome-extension/webpack.config.js, packages/vscode-extension/webpack.config.js, packages/mcp/src/sync.ts, packages/mcp/src/index.ts, packages/mcp/src/utils.ts, packages/core/src/index.ts, packages/core/src/types.ts, packages/mcp/src/config.ts, packages/core/src/context.ts, packages/mcp/src/handlers.ts, packages/mcp/src/snapshot.ts.
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.