# quillrag

> quillrag is an open-source MCP server for knowledge and memory. Single-binary local RAG MCP server in Rust for searching your own notes and docs. It is available on CLI, Linux, macOS, Windows and self-hosted.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** CLI, Linux, macOS, Windows, self-hosted
- **Repository:** https://github.com/ayush-yadav11/quillrag
- **GitHub stars:** 3
- **License:** MIT
- **MCP Registry name:** io.github.Ayush-yadav11/quillrag
- **Categories:** [knowledge and memory](https://ai-powered.com/categories/knowledge-and-memory)
- **Listing updated:** 2026-11-18

## About

The README says quillrag bundles the MiniLM embedding model inside one Rust binary, so indexing and search run on the local machine without Node, Python or a model download on first use. Retrieval combines dense cosine similarity with BM25 keyword matching, merged using Reciprocal Rank Fusion, which helps with exact tokens such as error codes and IDs.

It exposes rag_index, rag_search, rag_status and rag_clear as MCP tools, and the same engine is available as a command-line tool. Indexing is incremental: unchanged files are skipped and pruning is scoped to the directory passed in, so several source roots can share one data directory.

The README is explicit about limits. Dense search is an exact linear scan with no approximate index yet, so query latency grows with corpus size, and the README's measured figures stop at 1K chunks. It is aimed at personal knowledge bases, project docs and code rather than very large corpora.

## Features

- **Embedded MiniLM model:** The all-MiniLM-L6-v2 weights are compiled into the binary, so no separate model download is needed for the first query.
- **Hybrid retrieval:** Dense vector search and BM25 keyword search are fused with Reciprocal Rank Fusion.
- **Incremental indexing:** Unchanged files are skipped via content hash and only changed files are re-embedded.
- **Scoped pruning:** Deletions only affect files under the directory being indexed, so other roots in the same store are kept.
- **Single redb store:** Chunk text, vectors and metadata live in one redb file with atomic commits.
- **Lazy model loading:** The MCP handshake and status calls do not load the model, so the server starts quickly.

## Good for

- Searching a personal folder of markdown notes from inside an MCP-capable editor or chat client
- Giving an AI assistant retrieval over local project documentation and source code
- Keyword-plus-semantic lookup of error codes, IDs and exact tokens across local files
- Keeping document search fully offline for private material

## Tools

- `rag_index`: Incrementally index a directory or file.
- `rag_search`: Hybrid dense and BM25 retrieval returning ranked chunks with source paths.
- `rag_status`: Document and chunk counts, bytes indexed and file-type breakdown.
- `rag_clear`: Wipe the index.

## FAQ

### Does quillrag send data over the network?

The README states there is no network code path after installation, and that embeddings, storage and search all run locally.

### How large a corpus does it handle?

The README reports measured query times at 1K chunks (about 25 ms) and extrapolates higher; dense retrieval is an exact linear scan without an ANN index yet.

### Which file types are indexed by default?

Common text, markup, config and source formats such as md, txt, json, yaml, html, rs, py, js and go, extendable with -e.

### What license does it use?

The README says the project is MIT licensed, with the optional OCR models under CC BY-SA 4.0.

### Is quillrag free?

Yes. quillrag is open source, so you can use and self-host it for free.

### What are the best alternatives to quillrag?

The closest alternatives to quillrag are MiniRAG, rag-mcp, Ragmir, mcp-vl-msa-rs and Enterprise Internal Knowledge Base: Production-Ready RAG + MCP.

## Where this comes from

- Description: written from the project's README and its MCP Registry entry.

## Alternatives

- [MiniRAG](https://ai-powered.com/mcp/minirag): MiniRAG is an open-source MCP server for knowledge and memory.
- [rag-mcp](https://ai-powered.com/mcp/rag-mcp): rag-mcp is an open-source MCP server for knowledge and memory.
- [Ragmir](https://ai-powered.com/mcp/ragmir): Ragmir is an open-source MCP server for knowledge and memory.
- [mcp-vl-msa-rs](https://ai-powered.com/mcp/mcp-vl-msa-rs): mcp-vl-msa-rs is an open-source MCP server for knowledge and memory.
- [Enterprise Internal Knowledge Base: Production-Ready RAG + MCP](https://ai-powered.com/mcp/enterprise-internal-knowledge-base-production-ready-rag-mcp): Enterprise Internal Knowledge Base: Production-Ready RAG + MCP is an open-source MCP server for knowledge and memory.
- [RAG for your documents](https://ai-powered.com/mcp/rag-for-your-documents): RAG for your documents is an open-source MCP server for knowledge and memory.
- [Chaos Cypher](https://ai-powered.com/mcp/chaos-cypher): Chaos Cypher is an open-source MCP server for knowledge and memory.
- [cob-rag-mcp](https://ai-powered.com/mcp/cob-rag-mcp): cob-rag-mcp is an MCP server for knowledge and memory.

---
Source: https://ai-powered.com/mcp/quillrag
