wiki-rag-mcp

Automated RAG ingestion pipeline and MCP server for enterprise wikis powered by Qdrant and FastMCP via Docker Compose.

Architecture Overview

• Vector DB: Qdrant container running on port 6333.

• Ingestor: Python container (src/ingest.py) that fetches wiki pages, generates vector embeddings, writes to Qdrant, and exits.

• MCP Server: FastMCP container exposing the search_wiki tool over SSE on port 8080.

• Inference Engine: Your local host's LLM manager (e.g., LM Studio) hosting chat and embedding models.

Related MCP server: Markdown RAG MCP

Supported Wiki Platforms

The ingestor includes built-in API parsers for two major wiki platforms out of the box:

• MediaWiki: Public or self-hosted MediaWiki instances (e.g., Wikipedia, internal company wikis). Queries the standard api.php endpoints to fetch page lists and revision content.

• Atlassian Confluence: Enterprise Confluence Cloud or Data Center spaces. Connects via /api/v2/pages using a Bearer token (AUTH_TOKEN) to extract page storage format and sanitize HTML down to clean plain text.

Extending to Other Platforms

Because src/ingest.py uses a clean modular structure, adding support for other documentation engines (such as Notion, GitHub Wikis, Docusaurus, or GitBook) only requires adding a new fetch_ function that returns a list of title and content dictionaries:

def fetch_custom_wiki() -> list[dict]:
    # Fetch raw pages from target API
    return [{"title": "Page Title", "content": "Clean plain text content..."}]