Claude Context

Claude Context solves the problem of coding agents losing track of a large codebase’s structure by indexing it into a vector database and exposing semantic search over it via MCP, so agents like Claude Code, Cursor, and Gemini CLI can search code by meaning rather than exact string match. Instead of re-indexing the entire codebase on every change, it uses a Merkle-tree structure to detect exactly which files changed and update only those embeddings incrementally.

Built by Zilliz (the team behind the Milvus vector database), the project ships as an npm core package (@zilliz/claude-context-core), an MCP server package (@zilliz/claude-context-mcp), and a VS Code extension, giving multiple integration paths depending on whether you want IDE-native search or agent-facing MCP access. It supports OpenAI, Voyage AI, and other embedding providers rather than being locked to one.

MIT licensed, the project also points toward a sibling tool, memsearch, for persistent cross-session memory — positioning Claude Context specifically as the codebase-search layer rather than trying to also solve long-term agent memory.

What You Get

• Semantic (vector-embedding-based) code search over your entire codebase, exposed via MCP
• Merkle-tree-based incremental indexing that updates only changed files instead of re-indexing everything
• A VS Code extension and an MCP server package for different integration paths
• Support for multiple embedding providers (OpenAI, Voyage AI, and others) rather than one fixed backend

Common Use Cases

• Giving Claude Code, Cursor, or Gemini CLI semantic awareness of a large codebase instead of relying on exact-match grep
• Finding relevant code by describing what it does rather than knowing exact function or variable names
• Keeping a codebase’s searchable index up to date efficiently as files change, without full re-indexing
• Adding codebase-aware context retrieval to an agentic coding workflow via a standard MCP integration

Under The Hood

Architecture Claude Context separates the core indexing/embedding logic (@zilliz/claude-context-core) from the MCP server that exposes it to agents (@zilliz/claude-context-mcp) and a VS Code extension for IDE-native use, so the same underlying index can be consumed through multiple front-ends. The Merkle-tree structure over the codebase’s file tree is the key efficiency mechanism: it lets the system detect exactly which files changed since the last index without diffing or re-embedding the entire codebase.

Tech Stack TypeScript/Node.js (20+), using vector embeddings for semantic search (with OpenAI, Voyage AI, and other providers supported), MCP as the agent-facing protocol, and Merkle trees for incremental change detection. Built and maintained by Zilliz, the company behind the Milvus vector database.

Code Quality Very active development and a growing user base (12K+ stars) reflect real-world usage; the presence of separate core, MCP, and VS Code extension packages published independently to npm/VS Code Marketplace suggests attention to modular release management rather than a single monolithic package.

What Makes It Unique Most code-search tools for AI agents either rely on simple grep/regex or require re-embedding an entire codebase on every change; Claude Context’s Merkle-tree-based incremental indexing specifically targets keeping semantic search fresh and cheap to maintain as a codebase evolves, which matters for large, actively developed repositories.

Self-Hosting

Licensing Model MIT licensed — the core, MCP server, and VS Code extension are all open source with no license key.

Self-Hosting Restrictions None found for the indexing and MCP components; embedding provider usage (OpenAI, Voyage AI, etc.) requires your own API keys with those providers, which is a cost/dependency of the embedding step rather than a license gate in Claude Context itself.

License Key Required No.