AI导哈

知识图谱命令行工具

一个无头 MCP 服务器，使 Claude 等 AI 代理能够读写并推理科学命题知识图谱，用于研究知识管理和文献关联分析。

  • by az9713
  • Python
  • 本地部署
  • 1 星标
  • 更新于
在 GitHub 查看 az9713/knowledge-graph-cli

使用说明

Knowledge Graph CLI

Prototype / Experimental

This project is in early development. The interface is minimal and command-line based - expect a raw, developer-oriented experience rather than a polished user interface. Feedback and contributions welcome!

A headless MCP (Model Context Protocol) server that enables AI agents like Claude to read, write, and reason over a knowledge graph of scientific propositions. Think of it as a "second brain" for research that Claude can directly interact with.

What Does This Project Do?

Knowledge Graph CLI allows you to:

  • Store scientific claims as atomic units in a searchable knowledge graph
  • Find relationships between concepts using semantic (meaning-based) search
  • Trace intellectual lineages - how one idea led to another
  • Detect contradictions - find where new claims conflict with existing knowledge
  • Build a research graph - connect related ideas with explicit reasoning

Who Is This For?

  • Researchers who want Claude to help manage their knowledge base
  • Students building literature reviews or thesis research
  • Anyone who accumulates scientific knowledge and wants AI-assisted organization

Quick Start (5 Minutes)

Prerequisites

Before you begin, you need:

  1. Python 3.10 or higher - Download Python
  2. uv (Python package manager) - Install with: pip install uv
  3. OpenAI API Key - Get one here
  4. Claude Code or Claude Desktop - For interacting with the MCP server

Step 1: Clone or Download This Project

# If using git:
git clone <repository-url>
cd knowledge-graph-cli

# Or download and extract the ZIP file, then:
cd path/to/knowledge-graph-cli

Step 2: Set Your OpenAI API Key

Create a .env file in the project root:

# Windows (PowerShell)
echo "OPENAI_API_KEY=sk-your-actual-api-key-here" > .env

# Windows (Command Prompt)
echo OPENAI_API_KEY=sk-your-actual-api-key-here > .env

# Linux/Mac
echo "OPENAI_API_KEY=sk-your-actual-api-key-here" > .env

Replace sk-your-actual-api-key-here with your actual OpenAI API key.

Step 3: Install Dependencies

uv sync

This downloads and installs all required Python packages.

Step 4: Add to Claude Code

claude mcp add --transport stdio atomic-graph -- uv run python src/server.py

Step 5: Start Using!

Start a Claude Code session by running:

claude

Then try:

"Ingest a hypothesis that transformer attention mechanisms scale quadratically with sequence length"

Documentation

Document Description
Quick Start Guide 12 hands-on use cases to learn the system
Advanced Examples 15 power-user scenarios: discover hidden connections, generate hypotheses, uncover "new facts"
User Guide Complete user documentation
Developer Guide For developers extending the system
Architecture System design and data flow
CLAUDE.md Instructions for Claude Code

Available Tools

Once configured, Claude can use these tools:

Tool Purpose
ingest_hypothesis Add a new scientific claim to the knowledge graph
connect_propositions Create relationships between claims
semantic_search Find related concepts by meaning
find_scientific_lineage Trace how one idea connects to another
find_contradictions Check if a claim conflicts with existing knowledge
list_propositions Browse all stored claims
get_unit Get details of a specific claim
delete_unit Remove a claim and its connections (requires confirmation)

Project Structure

knowledge-graph-cli/
├── src/
│   ├── __init__.py       # Package marker
│   ├── server.py         # MCP server entry point (8 tools)
│   ├── model.py          # Pydantic data models
│   ├── graph_engine.py   # Knowledge graph logic (LanceDB + NetworkX)
│   └── persistence.py    # JSON persistence for relations and idempotency
├── tests/
│   ├── test_quick_start.py      # 12 use cases from Quick Start guide
│   └── test_advanced_examples.py # 8 advanced scenarios
├── data/                 # Stored knowledge (auto-created)
│   ├── atomic_graph.lance/  # LanceDB vector storage
│   ├── relations.json       # Graph edges
│   └── idempotency.json     # Idempotency cache
├── docs/                 # Documentation
├── .env                  # API keys (create this - see below)
├── .mcp.json            # MCP configuration
└── pyproject.toml       # Python dependencies

Testing

Knowledge Graph CLI includes comprehensive automated tests that verify all documented functionality:

# Run Quick Start use cases (12 tests)
uv run python tests/test_quick_start.py

# Run Advanced Examples (8 scenarios)
uv run python tests/test_advanced_examples.py

Test Results Summary

Quick Start Tests (12/12 passing):

  • Hypothesis ingestion with sources
  • Knowledge cluster building
  • Semantic connections
  • Natural language search
  • Unit exploration
  • Intellectual lineage tracing
  • Contradiction detection
  • Cross-domain knowledge linking

Advanced Examples Tests (8/8 passing):

  • Hidden connection discovery (Sleep → Creativity)
  • Hypothesis generation (Drug repurposing)
  • Research gap identification (Climate feedback)
  • Implicit contradiction uncovering (Economics)
  • Cross-domain innovation (Biology → Cybersecurity)
  • Emergent pattern discovery (Power laws)
  • "Aha!" moment generation (Urban resilience)
  • Historical idea archaeology (Internet origins)

Support

  • Issues: Report bugs or request features via GitHub Issues
  • Documentation: See the /docs folder for detailed guides

License

[Add your license here]

相关 MCP