Minirag MCP runs on your own machine — the client starts it, so there is no endpoint to ping. 224 installs a week from pypi. Last commit 7 Sep 2026.
Local-first RAG MCP server: hybrid search over a folder of your own documents
Today is the operative word: we check Minirag MCP every 15 minutes and re-read its code on every release. Watch it and you find out the day that stops being true.
This server runs on your own machine — install it with the package manager and the client starts it for you. Package name taken from the official registry entry.
claude mcp add minirag-mcp -- uvx minirag-mcp
{
"mcpServers": {
"minirag-mcp": {
"args": [
"minirag-mcp"
],
"command": "uvx"
}
}
}
[mcp_servers.minirag-mcp]
command = "uvx"
args = ["minirag-mcp"]
{
"mcpServers": {
"minirag-mcp": {
"args": [
"minirag-mcp"
],
"command": "uvx"
}
}
}
{
"mcpServers": {
"minirag-mcp": {
"args": [
"minirag-mcp"
],
"command": "uvx"
}
}
}
This one needs environment variables set before it will start:
BASE_DIR (One document root; also the security boundary for file access. Defaults to the process working directory. Ignored when BASE_DIRS is set.), BASE_DIRS (JSON array of document roots, e.g. ["/docs/a", "/docs/b"]. Takes precedence over BASE_DIR. An invalid value is a hard configuration error.), DB_PATH (LanceDB index directory. Defaults to <first root>/.minirag/lancedb, so each corpus gets its own index.), CACHE_DIR (Embedding model cache. Defaults to the platform user cache dir, so the ~220 MB model is downloaded once and shared.), MODEL_NAME (fastembed model id. Changing it makes existing vectors incompatible with new queries; pair with a new DB_PATH or a full re-ingest.), MAX_FILE_SIZE (Per-file size limit in bytes, enforced before parsing.), CHUNK_TOKEN_BUDGET (Retrieval-unit size in the embedding model's own tokens. Range 16-128; the ceiling is the model's trained sequence length.), RAG_HYBRID_WEIGHT (Keyword weight in the weighted RRF fusion, range 0.0-1.0. 0 is vector-only; higher values raise the BM25 contribution.), RAG_GROUPING (Result grouping filter. 'similar' keeps only the closest group; 'related' also keeps the next one. Unset means no grouping filter.), RAG_MAX_DISTANCE (Drop results whose vector distance exceeds this value. Lower is stricter. Unset means no distance filter.), RAG_MAX_FILES (Keep chunks from at most this many best-scoring files. Unset means no per-file filter.), RAG_INSTRUCTIONS_APPEND (Extra paragraph appended to the instructions the server hands the client at connect time, for corpus-specific guidance.), ALLOW_PRIVATE_URLS (Let ingest_url fetch hosts resolving to loopback, link-local, private, reserved or unspecified addresses. Off by default.).
The author declared them in the registry entry; get the values from the project itself.
Local RAG MCP server with hybrid search, PDF/DOCX support, and zero-config setup
Cross-session memory and hybrid RAG over your notes and docs. Local-first, with cited answers.
Local-first agentic RAG with citations - hybrid search, reranking, multimodal document retrieval
Local-first MCP tools for searching and drafting from your X/Twitter archive.
Local RAG MCP for markdown documentation. Retrieval only; host synthesizes answers.
Local FAISS vector database for RAG with document ingestion, semantic search, and MCP prompts.
Markdown-first long-term memory for AI agents. Hybrid BM25 + vector search, runs locally.
Shared memory + orchestration for your coding agents. Local-first MCP, vector RAG.
Answers built from our own checks of this server.