Devrag
Local vector search over your Markdown notes for Claude Code, cutting token use
- Pricing
- Open source
- Platforms
- CLI, macOS, Linux
- Version
- 1.2.0
- Listing updated
- today
Similar MCP servers
About Devrag
Devrag is an open-source MCP server for developer tools. It is available on CLI, macOS, Linux and Windows.
The README says DevRag exists to stop Claude Code from reading whole documents to find one fact. It indexes Markdown files from paths and glob patterns set in a config.json, stores embeddings in a local SQLite database using sqlite-vec, and returns only the relevant chunks when Claude asks a question in natural language.
It ships as a single Go binary that runs as an MCP server over stdio, and the same functions are available as CLI commands that print JSON by default. The embedding model, multilingual-e5-small, is downloaded on first run, and GPU use on macOS goes through CoreML with a CPU fallback.
The project is aimed at developers who keep large documentation sets in Git and want each machine to run its own local index. Search can be narrowed by directory or filename pattern. The README's benchmark, run on a MacBook Pro M2 with 100 files, reports about 40x fewer tokens than a full read.
Tools this server provides (5)
Taken from the project documentation; the running server may differ.
- search
- Semantic vector search with optional directory and file pattern filters.
- index_markdown
- Index a markdown file.
- list_documents
- List all indexed documents.
- delete_document
- Remove a document from the index.
- reindex_document
- Re-index a document.
What it does
- Semantic document search
- Finds relevant Markdown chunks from natural language queries using local vector embeddings.
- Glob-based indexing
- Indexes files from directory paths and recursive glob patterns defined in a config file.
- Filtered search
- Limits results by directory or filename pattern in addition to the query.
- Incremental sync
- Re-indexes changed files rather than the whole collection, and indexes on startup.
- CLI and MCP modes
- Runs as an MCP server for Claude Code or as standalone commands with JSON or text output.
- Local CPU or GPU embedding
- Uses automatic GPU or CPU detection, with CoreML on macOS and CPU fallback.
Good for
- Letting Claude Code find API specs or design notes without reading full files
- Sharing a team docs repository where each developer runs a local search index
- Searching Japanese and English notes from one local index
- Reducing context usage when working on large documentation sets
Pricing
Open source
Free and MIT licensed; runs locally.
Devrag FAQ
Which platforms does DevRag run on?
The README lists macOS, Linux and Windows, with prebuilt binaries for Apple Silicon and Intel macOS, Linux x64 and ARM64, and Windows x64.
Does DevRag need an internet connection?
It is needed on first run to download the embedding model. The README also lists proxy settings for environments that need them.
What file types does it index?
The README says it indexes Markdown (.md) files.
What license is DevRag released under?
The README states the MIT License.
Is Devrag free?
Yes. Devrag is open source, so you can use and self-host it for free.
What are the best alternatives to Devrag?
The closest alternatives to Devrag are scree, Deeprepo, Ragmir, indexio and Code-Graph-RAG.
Where this comes from
- Description: written from the project's README and its MCP Registry entry.