Claude Memory Full MCP - Hybrid Local RAG for AI Agents

Install Evidence

A portable RAG database for local docs.

Install the package, initialise a SQLite database, ingest documents, and expose hybrid retrieval to Claude through MCP.

npm package @gilligantechinc/claude-memory-rag

GitHub destination github.com/Gilligan-Tech-Inc/local-rag-mcp

npm install -g @gilligantechinc/claude-memory-rag
local-rag-mcp init
local-rag-mcp ingest ./docs --collection project
local-rag-mcp ingest ./contract.pdf --collection legal
local-rag-mcp search "how do I configure"
local-rag-mcp doctor

{
  "mcpServers": {
    "claude-memory-full-mcp": {
      "command": "local-rag-mcp",
      "env": {
        "LOCAL_RAG_DB": "/absolute/path/to/rag.db"
      }
    }
  }
}

Keyword search always works

Every ingest writes to SQLite FTS5, so search remains usable offline and without a model warm-up.

Vectors that scale

Run an Ollama embedding model and get semantic retrieval via a scalable sqlite-vec ANN index — with automatic fallback to a pure-JS search if the extension can't load. Same results either way.

No hidden cloud layer

No hosted vector database, no graph service, no account. The knowledge base is one local file you control.

New in v0.3.0

Scales further, ingests more.

— PDF ingest: rag_ingest_file now accepts .pdf alongside .md, .txt, and .json.
— Scalable vector search: an optional sqlite-vec ANN index replaces the in-process cosine scan for large corpora, with a transparent, identical-ranking fallback when it isn't available.

Claude Memory

Short authoritative project facts

Use for rules, deploy notes, decisions, preferences, and agent bootstrap context. Search mode: keyword/BM25 plus recency.

Open Lite page

Claude Memory Full MCP

Document and knowledge-base retrieval

Use for reference docs, meeting transcripts, technical specs, regulations, and larger corpora. Search mode: keyword, vector, or hybrid.

Open GitHub

Tool Surface

Eight tools for local document RAG.

Claude can ingest, search, retrieve, and maintain documents in the local RAG database.

Tool	Purpose
`rag_ingest_text`	Add text directly to the local RAG database.
`rag_ingest_file`	Ingest markdown, text, JSON, or PDF files.
`rag_search`	Search with keyword, vector, or hybrid retrieval.
`rag_get_document`	Fetch a document and its chunks by ID.
`rag_list_documents`	Browse ingested documents by collection or tag.
`rag_delete_document`	Remove one document and its chunks.
`rag_reindex`	Rebuild the FTS and vector indexes, and optionally fill missing embeddings.
`rag_stats`	Show document count, chunk count, embedding coverage, vector-index backend, and Ollama status.

01 / INGESTAdd documents

Files are chunked, indexed with FTS5, and optionally embedded through Ollama.

02 / SEARCHHybrid retrieval

BM25 and vector relevance combine when embeddings are present, with keyword fallback.

03 / CITEReturn sources

Results include document title, path, chunk, and score context for transparent agent use.