MCP Tooling Lab

You run a Model Context Protocol (MCP) server that exposes a small, secure set of AI-accessible capabilities—generating embeddings, indexing documents, and performing semantic search—backed by a vector store. This enables AI agents to retrieve relevant information from your data on demand, in a controlled and auditable way.

How to use

Use the MCP tool suite to empower AI agents with three core capabilities: generate embeddings for text, index documents into a vector store, and perform semantic search over indexed content. You can query the system to retrieve context, condition results with metadata filters, and supply grounded answers to user queries. When integrating with an MCP host such as Claude Desktop, you run the MCP server locally and connect the host to the server instance. The tools are designed to be used by MCP-compatible hosts and support realistic agent workflows that access enterprise data safely.

How to install

Prerequisites: you need Node.js and pnpm installed on your machine. You also need access to OpenAI embeddings and a Chroma vector store if you plan to index documents locally.

Step 1 — Install dependencies and set up the environment.

# Install dependencies
pnpm install

# Create environment file with required keys
# Example values shown as placeholders
OPENAI_API_KEY=your_openai_key
CHROMA_URL=http://localhost:8000
CHROMA_COLLECTION=mcp_tooling_lab
OPENAI_EMBED_MODEL=text-embedding-3-small

Step 2 — Start the local Chroma vector store (if you run it locally).

docker run --rm -p 8000:8000 chromadb/chroma

Step 3 — Run the MCP server locally.

pnpm dev

Example host configuration for Claude Desktop

If you connect to Claude Desktop, you can provide an explicit MCP server configuration so Claude can launch and communicate with your server in an isolated environment.

{
  "mcpServers": {
    "mcp_tooling_lab": {
      "command": "node",
      "args": [
        "/ABSOLUTE/PATH/TO/dist/server.js"
      ],
      "env": {
        "OPENAI_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

Example workflow

1. Index documents you want searchable, including any relevant metadata. 2. Query the system to retrieve context for a user question. 3. Let the host agent compose a grounded response using retrieved context.

Notes

This setup demonstrates a practical production-like pattern where models access real systems through a controlled, auditable interface. Adjust metadata filters and embedding/model options to fit your data domain and privacy requirements.

Available tools

embed_text

Generates OpenAI embeddings for an array of input strings.

index_documents

Embeds and indexes documents into a Chroma collection, including optional metadata.

vector_search

Performs semantic search over indexed documents using embeddings with optional metadata filters.