MD GRAPH
Quickstart
Vectorless Context Engine for Agentic AI

Kill the Vector Chunks.
Store Explicit Knowledge Graphs.

MdGraph is an open-source storage engine implementing Google's Open Knowledge Format (OKF). Ingest any format, compile with any LLM, and query traversable Markdown knowledge trees natively over MCP.

$ pip install mdgraph
Explore Architecture View on GitHub

Engineered for Complete Pluggability

Swap parsers, inference providers, and database indexes via clean configuration interfaces.

Universal Ingestion

Parse PDFs, HTML, PPTs, or audio transcripts into structured Markdown using Docling or Unstructured without losing structural layout or table integrity.

Pluggable LLM Compiler

Compile knowledge using any LLM—from local Llama models via Ollama to Claude, Gemini, or ChatGPT—powered by LiteLLM and strict Pydantic validation.

Git-Backed Truth

Markdown with YAML frontmatter is the absolute source of truth. Every concept modification is auditable, versioned, and committed via Libgit2.

Vectorless Graph Indexing

Ditch embeddings and cosine similarity. Uses SQLite FTS5 or PostgreSQL recursive CTEs to index cross-links and enable instant multi-hop topological queries.

Native MCP Server

Expose your entire knowledge graph over Model Context Protocol (FastMCP) so agents built in Google ADK, LangGraph, or CrewAI can consume context natively.

Google OKF Standard

Built natively around Google's Open Knowledge Format v0.1/v0.2 specification, turning ephemeral conversation logs into compounding knowledge bases.

System Architecture

Separating the knowledge compiler from the queryable graph cache.

1. Ingest

Docling / Parsers

2. Compile

LiteLLM + OKF Schema

3. Store

Git + Markdown Repo

Index Layer (Cache)

SQLite or PostgreSQL automatically indexes YAML frontmatter as JSON, Markdown bodies into FTS5 for text search, and cross-links into graph edge tables.

Interface Protocol

FastMCP exposes `get_concept()`, `search_knowledge()`, and `traverse_graph()` as standard tools for autonomous agent consumption.

Consumer Agents

External agents built with Google ADK, LangGraph, or custom clients navigate the graph deterministically without chunk ambiguity.

Strict Standards

Human-Readable Markdown.
Machine-Traversable Graphs.

Unlike black-box vector databases where data is trapped inside opaque float arrays, mdgraph keeps your core knowledge fully portable. Every document is regular Markdown readable by any text editor, versioned in Git, and instantly queryable.

  • Zero embedding generation costs or drift
  • Preserves multi-hop reasoning pathways
  • Ephemeral, rebuildable database indexes
concepts/remote_work_policy.md 
---
type: Policy
title: Remote Work Hardware Allowance
description: Home office equipment stipend rules
tags: [hr, hardware, remote]
status: stable
---

# Remote Work Hardware Allowance

Employees receive a yearly stipend managed by 
the [IT Department](./it_department.md).

## Approved Purchases
All equipment must be sourced via the 
[Vendor Portal](./vendor_portal.md).

Get Started in Minutes

Install the core package and start compiling knowledge graphs for your agent fleet.

# 1. Install mdgraph engine

pip install mdgraph

# 2. Start the local MCP server

mdgraph serve --config config.yaml

# 3. Connect from Google ADK, LangGraph, or Cursor

mcp://localhost:8000/sse