End-to-end example

Wire any MCP-capable agent to rag-hub-mcp, then drive everything — creating a KB, adding documents, and querying them — directly from your chat prompts. This example uses Claude Code, but it works the same with opencode, Cline, or any client exposing MCP tools.

1. Point the agent at the server

Add the MCP server to Claude Code's config (~/.claude.json or project .mcp.json). Using npx, the server starts on demand in stdio mode — no server to keep running:

{
  "mcpServers": {
    "rag-hub-mcp": {
      "command": "npx",
      "args": ["rag-hub-mcp"],
      "env": {
        "EMBEDDINGS_BASE_URL": "http://localhost:11434/v1",
        "KB_ROOT": "./kbs",
        "DB_PATH": "./rag.db"
      }
    }
  }
}

Restart the agent. The 8 rag_* tools (rag_list_kbs, rag_add_document, rag_search, …) are now available to your sessions.

2. Create a KB and add documents (via the agent)

> Using the rag-hub-mcp MCP tools, create a knowledge base called "infra"
  and add a document to it explaining how SSO authentication is set up.

The agent calls rag_add_document(kb="infra", path="sso.md", content="# SSO setup\n…"). The KB folder is created on the fly and the document is extracted, embedded and indexed.

> Also add a second document "network.md" about the DNS topology.

Now infra holds two documents and is ready to answer questions.

3. Query the index through your agent

> Using the rag-hub-mcp MCP tools, tell me: what is the DNS topology
  and how is SSO authentication configured?

The agent calls rag_search(query="…", kb="infra"), pulls the matching chunks with source citations, and answers from the indexed content — no need to browse the files yourself.

4. Check what's stored

> List the knowledge bases and the documents inside "infra".

This runs rag_list_kbs and rag_list_documents(kb="infra"), so you always know what has been indexed and when.

Tip: the same flow works headlessly via the REST API if you prefer curl over an agent.