Skip to content

Summary

Zilliz's MCP server for semantic code search: it indexes a whole repository into Milvus or Zilliz Cloud and hands coding agents only the relevant chunks.

Features

  • Hybrid BM25 plus dense-vector search over a whole codebase
  • AST-aware chunking with incremental re-indexing via Merkle trees
  • OpenAI, VoyageAI, Gemini or local Ollama embeddings
  • Stores vectors in self-hosted Milvus or Zilliz Cloud
  • Four tools: index, search, indexing status and clear index

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

  1. Open your agent's MCP configuration file.
  2. Merge the snippet below into it, filling in the environment variables with your own values.
  3. Restart the agent — the "Claude Context" tools will be available.
{
  "mcpServers": {
    "claude-context": {
      "command": "npx",
      "args": [
        "@zilliz/claude-context-mcp@latest"
      ],
      "env": {
        "OPENAI_API_KEY": "your-openai-api-key",
        "MILVUS_ADDRESS": "your-zilliz-cloud-public-endpoint",
        "MILVUS_TOKEN": "your-zilliz-cloud-api-key"
      }
    }
  }
}

Description

Claude Context is Zilliz's open-source MCP server for semantic code search. Rather than letting a coding agent grep its way through a large repository over several rounds, it indexes the codebase once into a vector database (a self-hosted Milvus or Zilliz Cloud) and answers natural-language questions such as "find the functions that handle user authentication" with the matching code chunks, so only relevant code lands in the context window.

Search is hybrid, combining BM25 keyword matching with dense vector similarity. Code is split along its abstract syntax tree, falling back to a character-based splitter, and re-indexing is incremental: a Merkle tree tracks which files changed. The server exposes four tools: index_codebase, search_code, get_indexing_status and clear_index. Embeddings can come from OpenAI, VoyageAI, Gemini or a local Ollama model, and the README lists TypeScript, JavaScript, Python, Java, C++, C#, Go, Rust, PHP, Ruby, Swift, Kotlin, Scala and Markdown as supported languages.

It runs locally through npx and needs Node.js 20 or newer, an embedding API key and a Milvus endpoint; Zilliz Cloud has a free tier. The README carries setup for Claude Code, Codex CLI, Gemini CLI, Qwen Code, Cursor, Windsurf, VS Code, Claude Desktop and more. The same engine ships as the @zilliz/claude-context-core package and a VS Code extension. MIT licensed.

Self-hosted documentation retrieval for coding agents — index any library's docs from the web, GitHub, npm, PyPI or local files and query them by exact version. An open-source Context7 alternative.

MCP: pg-aiguide

by Tiger Data

New

Tiger Data's hosted MCP server and skills for PostgreSQL: version-aware semantic search over the Postgres, TimescaleDB and PostGIS manuals.

MCP: Chroma

by Chroma

Chroma's official MCP server — build and query vector collections directly from an AI client.

MCP: Qdrant

by Qdrant

Qdrant's official MCP server — semantic memory for AI assistants backed by a real vector database.

Browse all MCP servers →