@vivantel/virage-mcp

MCP (Model Context Protocol) stdio server for Virage RAG. Expose your indexed knowledge base to AI assistants — search and inspect chunks directly from Claude Desktop, Claude Code, or any MCP-compatible client.

Features

• search — semantic vector search with similarity scores
• list_chunks — browse indexed chunks, filter by source file
• get_chunk — fetch a single chunk by content hash
• list_source_files — see all indexed files with chunk counts
• get_stats — index statistics (totals, embedding/upload status)

All tools are read-only — the server never modifies your index.

Installation

npm install @vivantel/virage-mcp

Usage

Requires a virage.config.json in your project (generated by virage init).

virage-mcp --config ./virage.config.json

Claude Desktop configuration

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "virage": {
      "command": "npx",
      "args": [
        "@vivantel/virage-mcp",
        "--config",
        "/absolute/path/to/your/virage.config.json"
      ]
    }
  }
}

Claude Code configuration

Run inside your project:

claude mcp add virage -- npx @vivantel/virage-mcp --config ./virage.config.json

Programmatic use

import { createMcpServer } from "@vivantel/virage-mcp";
import {
  loadConfig,
  VirageDb,
  defaultVirageDb,
} from "@vivantel/virage-core";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";

const cfg = await loadConfig("./virage.config.json");
const db = new VirageDb(defaultVirageDb());
await cfg.vectorStore.initialize();

const server = createMcpServer({
  db,
  embedder: cfg.embedder,
  vectorStore: cfg.vectorStore,
});
await server.connect(new StdioServerTransport());

Requirements

• Node.js ≥ 18
• A configured Virage project with virage.config.json and a populated .virage/virage.db

License

MIT © Vivantel