skills.rest
sandraschi
sandraschiCOMMUNITY
Explore all skills by this author →
GitHub

rag-expert

Provides shared vector search and document ingestion via LanceDB and sentence-transformers over HTTP.

2|Updated Sep 2, 2025
Search Tags:#semantic search#embeddings#MCP#rag#vector search#lancedb#sentence-transformers
One-click install
npx skills add https://github.com/sandraschi/local-llm-mcp --skill rag-expert-sandraschi
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: rag-expert
Source: https://github.com/sandraschi/local-llm-mcp/tree/main/skills/rag-expert
Command: npx skills add https://github.com/sandraschi/local-llm-mcp --skill rag-expert-sandraschi

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires httpx.

What problem does it solve? Lightweight fleet MCP repos often cannot justify shipping their own vector database and embedding stack. This Skill exposes a shared RAG-as-a-service endpoint so any repo can ingest and search documents over HTTP without adding LanceDB or sentence-transformers dependencies. ## Core Features & Use Cases - Shared Vector Search: Query a single LanceDB index embedded with all-MiniLM-L6-v2 across multiple repos and corpora. - HTTP Ingestion API: POST text with optional source and metadata to /api/v1/rag/ingest, then search with ranked scores via /api/v1/rag/search. - Use Case: A small fleet server needs to search arxiv papers. It fetches each paper, ingests the text through the shared endpoint, and retrieves matched sources with a single GET request instead of maintaining its own embedding pipeline. ## Quick Start Use the rag-expert skill to ingest a document into the shared RAG service and search it for transformer-related content.

Frequently Asked Questions about rag-expert

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I add vector search to a small MCP server without LanceDB?▼

Use the shared RAG endpoint at http://localhost:10833/api/v1/rag. POST text to /ingest and query /search with a query string and limit, receiving ranked results with text, source, metadata, and score.

How to ingest a web page into a LanceDB vector index?▼

Fetch the URL with httpx, cap the text at 50000 characters, then POST it to /api/v1/rag/ingest with the URL as the source field. The service embeds it with all-MiniLM-L6-v2 and stores it in LanceDB.

When should I not use a shared RAG service?▼

Avoid shared RAG when your repo holds private or sensitive data, needs custom chunking or metadata handling, or already has its own LanceDB index. In those cases keep a local RAG stack instead.

What embedding model does the shared RAG service use?▼

The service uses all-MiniLM-L6-v2 from sentence-transformers, with embeddings stored in LanceDB. You can confirm the active model and document count via the /api/v1/rag/status endpoint.

How do I clear all documents from the RAG index?▼

Send a DELETE request to /api/v1/rag/clear. The endpoint returns a success flag with cleared set to true, removing all ingested documents from the shared LanceDB index.

Related Skills

Explore More Software Engineering Skills→
mnazaalmnazaal

research-paper

Analyze research papers from PDF, URL, arXiv, DOI, or pasted text with structured critique.

Community
Intermediate
MerkyorLynnMerkyorLynn
37

rag-assistant

Retrieve relevant passages from an indexed local knowledge base and synthesize source-grounded answers.

Community
Intermediate
AdrianAVA9AdrianAVA9

health-checks

Configure health checks for .NET application dependencies with liveness and readiness endpoints.

Community
Intermediate
reason-machinesreason-machines
5

ai-data-engineering-rag-pipeline

Build local RAG pipelines with BM25 search, hierarchical chunking, and retrieval evaluation.

Official
Advanced