Claude Context

MCP servers

MCP server adding semantic code search over your whole codebase via a Milvus/Zilliz vector index.

by Zillizzilliztech/claude-context

Stars12.6k

Forks943

Watchers57

Open issues147

LicenseMIT

Last push3 months ago

Install

install

$ claude mcp add claude-context -e OPENAI_API_KEY=sk-your-key -e MILVUS_ADDRESS=your-zilliz-endpoint -e MILVUS_TOKEN=your-zilliz-key -- npx @zilliz/claude-context-mcp@latest

Update

update

$ claude mcp add claude-context -e OPENAI_API_KEY=sk-your-key -e MILVUS_ADDRESS=your-zilliz-endpoint -e MILVUS_TOKEN=your-zilliz-key -- npx @zilliz/claude-context-mcp@latest

npx-launched MCP server pinned to @latest: re-running add (or restarting the client) fetches the newest published version.

When to use it

Your codebase is too large to load directory-by-directory into context and you want semantic retrieval instead.
You want to cut token spend on large repos — Zilliz reports ~40% token reduction at equal retrieval quality.
You need cross-file code search inside Claude Code, Cursor, Codex, Gemini CLI, or other MCP clients.
You want incremental re-indexing (Merkle-tree diffing) so only changed files get re-embedded.

Commands & flags

index_codebase: Index a codebase directory for hybrid search (BM25 + dense vector).
search_code: Search the indexed codebase with natural-language queries via hybrid search.
get_indexing_status: Report indexing progress percentage and completion status for a codebase.
clear_index: Clear the search index for a specific codebase.
Index this codebase: Natural-language prompt that triggers the index_codebase tool in Claude Code.
Find functions that handle user authentication: Natural-language search prompt routed to search_code.

How to use it

This is an MCP server, not a skill or plugin. It needs a vector database (free tier on Zilliz Cloud) and an embedding API key (OpenAI, VoyageAI, Ollama, or Gemini). Register it once with the Claude Code CLI:

claude mcp add claude-context \
  -e OPENAI_API_KEY=sk-your-openai-api-key \
  -e MILVUS_ADDRESS=your-zilliz-cloud-public-endpoint \
  -e MILVUS_TOKEN=your-zilliz-cloud-api-key \
  -- npx @zilliz/claude-context-mcp@latest

Then, from your project directory, drive it with plain prompts:

Index this codebase
Check the indexing status
Find functions that handle user authentication

Requires Node.js >= 20. Works with Cursor, Codex CLI, Gemini CLI, Qwen Code, Windsurf, Void, and Claude Desktop via equivalent JSON/TOML config. Also ships a @zilliz/claude-context-core npm package and a "Semantic Code Search" VS Code extension for non-MCP use.

Why it earns its place

Built by Zilliz, the team behind Milvus, so the vector-search backbone is production-grade rather than hand-rolled. It is a Trendshift-listed, actively maintained MIT project with a published efficiency benchmark (~40% token reduction at equivalent retrieval quality), AST-based code chunking, and Merkle-tree incremental indexing. Unlike grep-based context tools, it gives agents true hybrid (BM25 + dense vector) recall over millions of lines while keeping per-request token cost low.

Try it

claude

> Index this codebase, then find all functions that handle user authentication and JWT validation.