Claude Context — Semantic Code Search MCP for AI Coding Agents

What does this agent do, and when should you use it?

Claude Context is an MCP plugin that adds semantic code search to Claude Code and other AI coding assistants, providing deep context from your entire codebase. It indexes code into a vector database and retrieves only relevant snippets, reducing token usage and cost while maintaining retrieval quality. The repository is a monorepo containing the core indexing engine (`@zilliz/claude-context-core`), the MCP server (`@zilliz/claude-context-mcp`), and a VS Code extension. It supports hybrid search (BM25 + dense vectors), incremental indexing via Merkle trees, and AST-based code chunking. Deployment requires an OpenAI embedding API key and a Milvus or Zilliz Cloud vector database. The MCP server follows the standard MCP protocol with stdio transport, integrating with multiple clients such as Claude Code, Codex CLI, Cursor, and more.

Claude Context takes a codebase directory as input, splits files into chunks, generates embeddings using OpenAI models, and stores them in a Milvus/Zilliz Cloud vector database. It exposes MCP tools: index_codebase, search_code, clear_index, and get_indexing_status. On search, it performs hybrid retrieval (BM25 + dense vectors) to return relevant code snippets with file paths, line numbers, and content. Core classes include Context, MilvusVectorDatabase, and OpenAIEmbedding. The MCP server runs via npx @zilliz/claude-context-mcp, and configuration uses environment variables OPENAI_API_KEY, MILVUS_ADDRESS, and MILVUS_TOKEN.

1. A developer working on a large or unfamiliar codebase wants to quickly locate functions implementing specific behavior without manual grep.
2. An engineering team wants to reduce token costs when using Claude Code on big repos by loading only relevant code into context.
3. A team using multiple AI coding assistants (e.g., Cursor, Gemini CLI) needs a standardized MCP-based semantic search across tools.
4. A developer maintaining a fast-moving codebase needs incremental indexing to keep the semantic index up to date efficiently.
5. A VS Code user wants to search code using natural language within the IDE and see results with line numbers directly.

What are this agent's strengths and limitations?

How do you install or deploy this agent?

Prerequisites: Node.js >= 20; obtain an OpenAI API key; set up a Zilliz Cloud instance (or Milvus) and get the public endpoint and API token.

For Claude Code, run:

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

For other clients (Codex CLI, Gemini CLI, Cursor, etc.), see the README sections with configuration snippets.

How do you use this agent?

1. Start Claude Code in your project directory (cd your-project && claude).
2. Type Index this codebase to index.
3. Check status with Check the indexing status.
4. Ask questions like Find functions that handle user authentication.

You can also use the core package programmatically in TypeScript:

import { Context, MilvusVectorDatabase, OpenAIEmbedding } from '@zilliz/claude-context-core';
const embedding = new OpenAIEmbedding({ apiKey: 'your-openai-api-key', model: 'text-embedding-3-small' });
const vectorDatabase = new MilvusVectorDatabase({ address: 'your-zilliz-cloud-public-endpoint', token: 'your-zilliz-cloud-api-key' });
const context = new Context({ embedding, vectorDatabase });
const stats = await context.indexCodebase('./your-project', (progress) => {});
const results = await context.semanticSearch('./your-project', 'vector database operations', 5);

How does this agent compare with similar options?

FAQ