Metadata-Version: 2.4
Name: rbt-mcp-server
Version: 0.3.2
Summary: MCP Server for Graphiti memory and document chunking with intelligent Markdown processing.
Project-URL: Homepage, https://github.com/yourusername/KnowledgeSmith
Project-URL: Repository, https://github.com/yourusername/KnowledgeSmith
Project-URL: Issues, https://github.com/yourusername/KnowledgeSmith/issues
Author-email: Devin Lai <leo7nel23@gmail.com>
License: MIT
Keywords: document-editor,llm,mcp,rbt,token-optimization
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Requires-Python: >=3.10
Requires-Dist: graphiti-core>=0.3.0
Requires-Dist: jsonschema>=4.0.0
Requires-Dist: mcp>=1.0.0
Requires-Dist: pyyaml>=6.0.0
Requires-Dist: referencing>=0.30.0
Requires-Dist: rpds-py>=0.20.0
Provides-Extra: dev
Requires-Dist: pytest-cov>=4.0.0; extra == 'dev'
Requires-Dist: pytest>=7.0.0; extra == 'dev'
Description-Content-Type: text/markdown

# KnowledgeSmith MCP Server

MCP Server for **Graphiti memory and document chunking**. Previously included RBT document editing tools (now archived).

## 📦 Archive Notice

### RBT Document Editor Tools (Archived 2025-10-09)

The RBT document editing功能已於 2025-10-09 封存，改用原生 Claude Code Read/Edit/Write 工具以降低維護成本和 token 使用。

**封存內容：**
- document_service.py - 文件服務
- document_parser.py - 文件解析器
- 11 個 editor MCP 工具（get_outline, read_content, update_block 等）
- templates/ - 文件模板
- cache.py - 文件快取

**保留功能：**
- ✅ chunking/ - 文件分塊與同步功能
- ✅ graphiti_tools.py - Graphiti 記憶體功能（8 個工具）

**如何恢復封存的代碼：**
```bash
# 查看封存版本
git show v-with-editor

# 恢復特定檔案
git checkout v-with-editor -- rbt_mcp_server/document_service.py

# 或建立分支使用完整封存版本
git checkout -b restore-editor v-with-editor
```

## 🎯 Current Features

### Graphiti Knowledge Graph Integration
- **Intelligent Chunking**: Automatically split documents into semantic chunks based on document structure (sections for RBT, H3 headings for Markdown)
- **Incremental Sync**: Only update changed chunks, preserving unchanged content
- **Neo4j Backend**: Store document chunks as episodes in Graphiti knowledge graph
- **graphiti-memory Compatible**: Drop-in replacement with same search_nodes/search_facts API
- **8 MCP Tools**: add_document, search_memory_nodes, search_memory_facts, get_episodes, delete_episode, get_entity_edge, delete_entity_edge, clear_graph

## 📦 Installation

### Prerequisites

**1. Setup Neo4j Database**

Using Docker (recommended):
```bash
docker run \
  -p 7474:7474 \
  -p 7687:7687 \
  -e NEO4J_AUTH=neo4j/your-password \
  --name neo4j \
  neo4j:latest
```

Verify at: http://localhost:7474

**2. Get OpenAI API Key**

Required for Graphiti embeddings and graph operations.

### Install MCP Server

**Option 1: Install from source (uv)**

```bash
# Clone repository
git clone https://github.com/yourusername/KnowledgeSmith.git
cd KnowledgeSmith

# Install with uv
uv pip install -e .
```

**Option 2: Direct installation**

```bash
uv pip install rbt-mcp-server
```

## 🚀 Quick Start

### 1. Configure Claude Desktop

Add to `~/Library/Application Support/Claude/claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "graphiti-memory-server": {
      "type": "stdio",
      "command": "rbt-mcp-server",
      "env": {
        "RBT_ROOT_DIR": "/path/to/your/document/root",
        "NEO4J_URI": "bolt://localhost:7687",
        "NEO4J_USER": "neo4j",
        "NEO4J_PASSWORD": "your-password",
        "OPENAI_API_KEY": "your-openai-api-key"
      }
    }
  }
}
```

**Required Environment Variables**:
- `RBT_ROOT_DIR`: Root directory for document comparison (required for add_document tool)
- `NEO4J_URI`, `NEO4J_USER`, `NEO4J_PASSWORD`: Neo4j database connection
- `OPENAI_API_KEY`: OpenAI API key for Graphiti embeddings

Or use full uv command:

```json
{
  "mcpServers": {
    "graphiti-memory-server": {
      "type": "stdio",
      "command": "uv",
      "args": ["run", "rbt-mcp-server"],
      "env": {
        "RBT_ROOT_DIR": "/path/to/your/document/root",
        "NEO4J_URI": "bolt://localhost:7687",
        "NEO4J_USER": "neo4j",
        "NEO4J_PASSWORD": "your-password",
        "OPENAI_API_KEY": "your-openai-api-key"
      }
    }
  }
}
```

### 2. Set Environment Variables (Optional - if not using Claude Desktop)

```bash
# Required for add_document tool
export RBT_ROOT_DIR=/path/to/your/document/root

# Required for Graphiti integration
export NEO4J_URI=bolt://localhost:7687
export NEO4J_USER=neo4j
export NEO4J_PASSWORD=your-password
export OPENAI_API_KEY=your-openai-api-key
```

### 3. Test the Server

```bash
rbt-mcp-server
```

## 📚 Available MCP Tools

### Document Management

1. **add_document** - Sync documents to knowledge graph with automatic chunking
   - Supports Markdown (chunked by H3 headings) and RBT documents
   - Incremental sync: only updates changed chunks

### Knowledge Graph Query

2. **search_memory_nodes** - Search knowledge graph nodes (entities, preferences, procedures)
3. **search_memory_facts** - Search knowledge graph facts (relationships)
4. **get_episodes** - Retrieve recent memory episodes

### Data Management

5. **delete_episode** - Delete specific episode
6. **get_entity_edge** - Get entity relationship edge by UUID
7. **delete_entity_edge** - Delete entity relationship edge
8. **clear_graph** - Clear all data from knowledge graph (⚠️ irreversible)

## 🔗 Graphiti Integration Usage

### Adding Documents to Knowledge Graph

**General Markdown Documents:**
```python
add_document(
    new_file_path="/absolute/path/to/document.md",
    project_id="my-project",
    file_path="docs/guide.md"  # relative path for general docs
)
```

**RBT Documents (REQ/BP/TASK):**
```python
add_document(
    new_file_path="/absolute/path/to/TASK-001.md",
    project_id="knowledge-smith",
    feature_id="my-feature",
    rbt_type="TASK",
    file_path="001"  # task number for TASK documents
)
```

### Searching Knowledge

```python
# Search for nodes (entities, preferences, procedures)
results = await search_nodes(
    query="documentation preferences",
    group_ids=["knowledge-smith"],
    entity="Preference",
    max_nodes=10
)

# Search for facts (relationships)
facts = await search_facts(
    query="task dependencies",
    group_ids=["knowledge-smith"],
    max_facts=10
)
```

### Difference from graphiti-memory MCP

This MCP server extends the original `graphiti-memory` MCP with document chunking capabilities:

- **Original graphiti-memory**: Stores entire documents as single episodes
- **This MCP (graphiti-chunk-mcp)**: Automatically chunks documents into semantic sections
  - RBT documents: Split by section (sec-*)
  - Markdown documents: Split by H3 headings (###)
  - Incremental updates: Only sync changed chunks

**API Compatibility**: All `search_nodes`, `search_facts`, `get_episodes` functions maintain the same interface as graphiti-memory.

## 📖 Documentation

- **[Markdown Writing Guide](docs/guides/markdown-writing-guide.md)** - How to write Markdown documents that work well with the chunker
- **[MCP Installation Guide](docs/guides/mcp-installation-guide.md)** - Detailed installation and usage instructions

## 🧪 Development

Install development dependencies:
```bash
uv sync --dev
```

Run tests:
```bash
RBT_ROOT_DIR=/test/root uv run pytest -v
```

Test coverage:
```bash
RBT_ROOT_DIR=/test/root uv run pytest --cov=rbt_mcp_server --cov-report=html
```

## 📝 License

MIT License

## 🤝 Contributing

Contributions welcome! Please open an issue or submit a pull request.
