Claude Context

You set up the Claude Context MCP Server to index and semantically search codebases. This MCP server lets you connect Claude Context to your favorite coding assistants, enabling natural-language queries to locate code across large projects with fast, hybrid search.

How to use

Use an MCP client that supports standard MCP servers. Add the Claude Context MCP server to your client configuration so you can index codebases and run semantic searches. You can index an entire repository or directory, then query with natural language like “find usages of function X” or “show code snippets related to Y”. The server combines traditional keyword search with dense vector representations for accurate results, and supports incremental indexing so changes are reflected efficiently.

How to install

Prerequisites: Install Node.js and npm, or ensure your runtime supports npm commands. Have a Milvus vector store available (local or cloud) and an embedding provider key from one of the supported providers (OpenAI, VoyageAI, Gemini, or Ollama). Prepare your Milvus token as an environment variable when running the MCP server.

Step 1: Create an environment configuration for the embedding provider you plan to use. For example, choose OpenAI and supply your API key. You will also set optional embedding models if desired.

Step 2: Install the MCP server package in your project or workspace using npm or your preferred package manager.

Step 3: Start the MCP server so it becomes available to MCP clients. Ensure the server can reach Milvus and the embedding provider endpoints.

Configuration and usage notes

Environment variables you may configure include the embedding provider, API keys, and Milvus token. You can also adjust indexing behavior, such as batch size, custom file extensions, and ignore patterns. The server accepts both http-based MCP connections to a remote server and stdio-based local execution via command invocations.

Key environment variables you will see in examples include OPENAI_API_KEY for the selected provider and MILVUS_TOKEN for Milvus access. You may also specify EMBEDDING_PROVIDER to select among OpenAI, VoyageAI, Gemini, and Ollama, and optional EMBEDDING_MODEL to pick a specific embedding model.

Example of a standard stdio-based configuration to run the MCP server locally using npm-like tooling is shown below. This configuration runs the MCP server via npx and loads the Claude Context MCP package.

{
  "mcpServers": {
    "claude_context": {
      "type": "stdio",
      "name": "claude_context",
      "command": "npx",
      "args": ["-y", "@zilliz/claude-context-mcp@latest"],
      "env": {
        "OPENAI_API_KEY": "your-openai-api-key",
        "MILVUS_TOKEN": "your-zilliz-cloud-api-key"
      }
    }
  }
}

Notes on configuration options

Embedding provider configuration supports multiple providers. You specify the provider, API keys, and optional models per provider. You can also adjust how many embeddings are batched together when indexing, and configure which file types to include or ignore during indexing.

Security considerations

Treat API keys and Milvus tokens as sensitive information. Do not commit keys to source control. Use environment-based configuration in deployment and restrict access to the MCP server endpoints.

Troubleshooting tips

If indexing or querying fails, verify network access to Milvus and the embedding provider, confirm that the API keys are valid, and check that the MCP server process is running and reachable by your MCP client. Review any startup timeout settings if your environment is slow to initialize.

Examples and quick references

To get started quickly, set EMBEDDING_PROVIDER to OpenAI and provide your OPENAI_API_KEY, then run the MCP server with the standard npx-based command shown above. You can later switch to VoyageAI, Gemini, or Ollama by updating EMBEDDING_PROVIDER and the corresponding API keys.

Available tools

index_codebase

Index a codebase directory for hybrid search (BM25 + dense vector). It accepts a path, optional force, split options, and custom extensions/ignore patterns.

search_code

Search the indexed codebase with natural language queries using hybrid search, returning ranked results.

clear_index

Clear the search index for a specific codebase path.

get_indexing_status

Show the current indexing progress or completion status for a codebase.