nkapila6/mcp-local-rag

Overview

A locally run Model Context Protocol (MCP) server that performs a primitive RAG-like web search without relying on any external APIs. It indexes and retrieves content using local resources only, emphasizing simplicity and offline operation.

Best for

Best for
Developers exploring local RAG and MCP with a no-API, keep-it-simple approach

Use cases

• Building a private, offline retrieval-augmented generation workflow
• Prototyping a local search tool for personal documents or intranet content
• Experimenting with MCP and RAG concepts without cloud dependencies

How to use

Tools exposed

• deep_research
• deep_research_google
• deep_research_ddgs
• rag_search_ddgs
• rag_search_google

Tested with

Claude Desktop, Cursor, Goose

Example client config

{\n  "mcpServers": {\n    "mcp-local-rag":{\n      "command": "uvx",\n        "args": [\n          "--python=3.10",\n          "--from",\n          "git+https://github.com/nkapila6/mcp-local-rag",\n          "mcp-local-rag"\n        ]\n      }\n  }\n}

Notes

A locally run Model Context Protocol (MCP) server that performs a primitive RAG-like web search without relying on any external APIs. It indexes and retrieves content using local resources only, emphasizing simplicity and offline operation.

126 stars on GitHub. Last updated 2026-05-18. Licensed MIT.

Use cases

• Building a private, offline retrieval-augmented generation workflow
• Prototyping a local search tool for personal documents or intranet content
• Experimenting with MCP and RAG concepts without cloud dependencies

Pros

• No external API keys or costs required
• Fully offline, preserving data privacy
• Lightweight Python implementation easy to modify

Cons

• Described as ‘primitive’, so likely limited indexing and retrieval quality
• May lack advanced RAG features like chunking, embedding optimization, or ranking
• Small community (126 stars) and minimal documentation beyond the repo

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.