Happy slice of mcp pizzaMCP.pizza

mcp-local-rag

MCP.Pizza Chef: nkapila6

Five tools handle quick lookups and longer research runs across DuckDuckGo, Google, Bing, Brave, Wikipedia, Yahoo, Yandex, Mojeek and Grokipedia. A small ranking model ships inside the package and scores results on your computer, so that part really is local, but the searches themselves and the page downloads still travel out to the internet. One caution: picking the privacy-first DuckDuckGo option does not pin the search to DuckDuckGo, because the underlying library quietly falls back to choosing engines for you.

Web/Research

Use This MCP server To

Look up current information without paying for a search key Research one topic across several search engines at once Pull the text of the best results into my chat Check a claim against Wikipedia and a couple of engines Compare what different search engines say about a company

README

mcp-local-rag

"primitive" RAG-like web search model context protocol (MCP) server that runs locally. ✨ no APIs ✨

A RAG-based web search and deep research model context protocol (MCP) server that runs entirely locally. Features multi-engine research across 9+ search backends with semantic similarity ranking, and requires no API keys.

Open in GitHub Codespaces

Add MCP Server mcp-local-rag to LM Studio

Ask DeepWiki

%%{init: {'theme': 'base'}}%%
flowchart TD
    A[User] -->|1.Submits LLM Query| B[Language Model]
    B -->|2.Sends Query| C[mcp-local-rag Tool]
    
    subgraph mcp-local-rag Processing
    C -->|Search DuckDuckGo| D[Fetch 10 search results]
    D -->|Fetch Embeddings| E[Embeddings from Google's MediaPipe Text Embedder]
    E -->|Compute Similarity| F[Rank Entries Against Query]
    F -->|Select top k results| G[Context Extraction from URL]
    end
    
    G -->|Returns Markdown from HTML content| B
    B -->|3.Generated response with context| H[Final LLM Output]
    H -->|5.Present result to user| A

    classDef default stroke:#333,stroke-width:2px;
    classDef process stroke:#333,stroke-width:2px;
    classDef input stroke:#333,stroke-width:2px;
    classDef output stroke:#333,stroke-width:2px;

    class A input;
    class B,C process;
    class G output;
Loading

Features

mcp-local-rag FAQ

Do I need to sign up or pay?
No. Nothing here asks for an account or a key, and the ranking model is bundled in the download rather than fetched from a paid service. That is the main reason to choose it over the hosted search connectors.
Is it actually private?
Partly, and the claim needs trimming. The scoring runs on your machine, but every search and every page fetch leaves your computer as normal. The DuckDuckGo-only tools are also not enforced in the code: the search library defaults to picking engines automatically, so your query may reach Google or others anyway.
How hard is it to set up?
Fiddly. You need either Docker or the uv tool installed before the config snippet does anything, and the first run pulls down a sizeable machine-learning package. Once that is done it starts on its own.
Which apps does it work in?
Claude Desktop, Cursor, Goose and LM Studio are all documented, and LM Studio has a one-click install link.
Can it change anything on my computer?
No. It only reads. It searches, downloads the top pages, trims each to about ten thousand characters and hands the text back. Nothing is written, deleted or executed.
Will searches always work?
Not always. It reads free public search pages rather than a paid service, so engines sometimes rate-limit or block it. When one engine fails the tool moves on quietly, which can make a research run come back thinner than expected.
Can I use this to research a topic and get sources?
Yes. Ask for deep research on a subject and it will run several related searches, rank what comes back, remove duplicates and return the text with links you can check.