Skip to main content

RAG Over Enterprise Content with Viglet Turing and MCP

· 3 min read
Alexandre Oliveira
Alexandre Oliveira
Creator & Lead Developer, Viglet Turing

Retrieval-Augmented Generation (RAG) is only as good as what it retrieves. If the retrieval layer is a throwaway vector index built for one chatbot, you end up rebuilding it for every new use case. The better pattern: make your enterprise search index the retrieval layer, and expose it to AI through the Model Context Protocol (MCP) so any agent can use it as a tool.

That's what Viglet Turing does: your indexed Adobe AEM, WordPress, database, and file content becomes both a search experience and a grounded RAG/agent capability, on your own infrastructure.

RAG and search are the same retrieval problem​

A search query and a RAG prompt both need the same thing: the most relevant passages from your content. Turing already maintains that index (faceted, multi-language, semantically ranked), so RAG reuses it instead of standing up a parallel store:

Built-in RAG over your content​

With content indexed (see the AEM guide), turning on RAG gives grounded answers with citations, using the LLM of your choice: OpenAI, Ollama, Anthropic, or Gemini:

curl "http://localhost:2700/api/sn/wknd-search/chat?q=What+is+our+refund+policy"

Answers are constrained to retrieved passages, so the model cites your content rather than hallucinating, and with Ollama the whole loop stays on your infrastructure.

MCP: your search as a tool for any agent​

The Model Context Protocol is an open standard that lets AI models call tools on external servers. Turing uses MCP in two directions:

  1. As an MCP client: an AI Agent in Turing can connect to external MCP servers (over HTTP or stdio) and use their tools alongside its 27 built-in tools. That can be a company knowledge system, a proprietary API, or any public MCP server, with no custom code inside Turing.
  2. As a retrieval capability: your enterprise index is exposed through tools like list_sites, get_site_fields, and search_site, so an agent can discover and search your content autonomously.

This is what turns "a search box" into "a retrieval backend any AI can build on": the same index powers your site search, your RAG chatbot, and external agents.

  • One index, many consumers: search UI, RAG chat, and MCP tools all read the same content. No duplicate pipelines to keep in sync.
  • Governed retrieval: facets, locale, and access scoping that already exist for search apply to what RAG and agents can retrieve.
  • Self-hosted and open: Apache 2.0, your LLM, your infrastructure; no content shipped to a third-party RAG service.

Next steps​

Viglet Turing is open-source enterprise search with semantic navigation, generative AI, and MCP support. Index your enterprise content once and use it as search, grounded RAG, and a tool for AI agents, on your own infrastructure.

ON THIS PAGE