Preconditions

Vector indexes configuration in Mongo Atlas

	{
	  "fields": [
		{
		  "numDimensions": 1536,
		  "path": "embedding",
		  "similarity": "cosine",
		  "type": "vector"
		},
		{
		  "path": "repo",
		  "type": "filter"
		},
		{
		  "path": "team",
		  "type": "filter"
		},
		{
		  "path": "name",
		  "type": "filter"
		},
		{
		  "numDimensions": 1536,
		  "path": "docsEmbedding",
		  "similarity": "cosine",
		  "type": "vector"
		}
	  ]
	}

that defines 2 vectors fields:

• embedding - consists of general application details

Includes the following fields: repo (github link), name (application name), team, summary, owners (codeowners), responsibilities, tags, dependencies

• docsEmbedding - consists of summary for found documentation

Application components

MongoDB Atlas

Represents a vector storage for RAG.

MCP server

Provides RAG functionality to AI models.

ReportAISummary Mcp Client

Provides a way to test MCP server locally.

How to use RAG

1. Run API application via Scalar UI.
2. Run ReportAISummary.Mcp.Client shell based application as MCP client.
3. Attach created MCP server to any AI aware env like cursor AI, Visual Studio, Github copilot and so on.

Supported Functionality

1. Repository Parsing: Automatically parses GitHub repositories listed in supported-repos.json and stores the extracted data in vector storage for efficient semantic querying.

2. MCP Integration: Provides Model Context Protocol (MCP)–based endpoints to ask questions and retrieve insights from the previously indexed repository data.

Future plans

1. Make indexing github repository data fully async.
2. Make RAG querying more precise including better score values.
3. Clarify components and env. Is MongoDB the best option here? Deploy to target ecosystem.
4. Add tests.