grp-kb 0.1.0

Dual-model semantic search MCP server for an Acumatica KB (general English + multilingual GRP-manual domains), no Obsidian/Smart Connections required

grp-kb

Dual-model semantic search MCP server for an Acumatica knowledge base. Two independently-embedded domains over one vault of markdown notes:

• general — everything except GRP_Manual*.md, embedded with TaylorAI/bge-micro-v2
• grp — GRP_Manual*.md only, embedded with paraphrase-multilingual-MiniLM-L12-v2 (the GRP manuals are in Bahasa Malaysia — bge-micro-v2 alone tested noticeably worse there than a multilingual model)

No Obsidian, no Smart Connections plugin, anywhere in this pipeline — indexes are built directly from plain .md files with sentence-transformers.

This package ships code only. It needs a real, pre-built vault + index pair to do anything useful — either build your own (see below) or get one shared from someone who already has it (just the .md vault + two kb_index.npy / kb_meta.json pairs, no re-embedding needed on the receiving end).

Build an index

Run twice — once per domain — pointing at separate KB_MCP_INDEX_DIR folders:

# general domain
export KB_VAULT_DIR=/path/to/vault
export KB_FILE_GLOB="*.md"
export KB_EMBED_MODEL=TaylorAI/bge-micro-v2
export KB_MCP_INDEX_DIR=/path/to/index
python -m grp_kb.build_index

# grp domain
export KB_VAULT_DIR=/path/to/vault
export KB_FILE_GLOB="GRP_Manual*.md"
export KB_EMBED_MODEL=sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2
export KB_MCP_INDEX_DIR=/path/to/index_minilm_grp
python -m grp_kb.build_index

Run the server

export KB_VAULT_DIR=/path/to/vault
export KB_GENERAL_INDEX_DIR=/path/to/index
export KB_GRP_INDEX_DIR=/path/to/index_minilm_grp
grp-kb

Register with Claude Code

claude mcp add grp-kb -s user \
  -e KB_VAULT_DIR=/path/to/vault \
  -e KB_GENERAL_INDEX_DIR=/path/to/index \
  -e KB_GRP_INDEX_DIR=/path/to/index_minilm_grp \
  -- grp-kb

Tools

• search_kb(query, top_k=10, source_only=False, guide_filter="") — returns a JSON array of results tagged source: "general" or "grp". top_k applies per domain, so a call can return up to 2x that many results total.
• read_kb_file(path) — full markdown content by relative path (as returned by search_kb).