shuruheel/mcp-engram

部署与 DevOps

本页由 AI 翻译，技术细节请以原文为准。查看原文

24 stars 5 forks 质量 79 趋势 18MIT

Repository Install

一个复杂的知识图谱记忆系统，存储具有丰富语义结构的相互关联信息。

概览

一个模型上下文协议（MCP）服务器，用于构建和探索受认知神经科学启发的知识图谱。基于人类思维结构，建模思维、情感、推理和情节记忆。支持SQLite——零基础设施，磁盘单文件，离线运行。服务器默认将数据库存储在 ~/.mcp-engram/knowledge.db。无需外部服务。 MCP服务器展示了8个工具，使LLM能够构建、搜索和遍历知识图谱：每个工具都会声明一个输入模式（已验证）、一个输出模式和行为注释（只读与写），并返回机器可读的结构化内容以及一个JSON文本块。写入：重新创建现有节点或边会更新你提供的字段，并保留未提供的字段。服务器还会暴露MCP提示——一个包含整体工具使用工作流程的系统提示，以及每个工具名称下的每个工具使用指南。

概述（AI 翻译）

一个模型上下文协议（MCP）服务器，用于构建和探索受认知神经科学启发的知识图谱。基于人类思维结构，建模思维、情感、推理和情节记忆。支持SQLite——零基础设施，磁盘单文件，离线运行。

快速启动

或者全局安装：

服务器默认将数据库存储在~/.mcp-engram/knowledge.db。无需外部服务。

建筑

工作原理

MCP服务器展示了8个工具，使LLM能够构建、搜索和遍历知识图谱：

工具	目的
`search_nodes`	跨所有节点的全文搜索（FTS5）
`explore_context`	加权图绕给定节点遍历
`create_nodes`	创建或合并更新任意类型的节点
`create_relations`	创建带有上下文、权重和信心的边
`add_sources`	记录来源（源节点 + DERIVED_FROM链接）
`get_temporal_sequence`	按时间顺序排列（下一步、之前、原因）
`create_reasoning_chain`	构建结构化多步推理
`get_reasoning_chain`	按名称或主题检索推理链

本页由 AI 翻译，技术细节请以原文为准。

README

以下为原文 README 节选，完整内容请查看原文。

MCP Engram — Cognitive Knowledge Graph

A Model Context Protocol (MCP) server for building and exploring a cognitive neuroscience-inspired knowledge graph. Models thought, emotion, reasoning, and episodic memory after the structure of the human mind. Backed by SQLite — zero infrastructure, single file on disk, works offline.

Quick Start

npx mcp-engram

Or install globally:

npm install -g mcp-engram
mcp-engram

The server stores its database at ~/.mcp-engram/knowledge.db by default. No external services required.

Architecture

How It Works

The MCP server exposes 8 tools that let an LLM build, search, and traverse a knowledge graph:

Tool	Purpose
`search_nodes`	Full-text search across all nodes (FTS5)
`explore_context`	Weighted graph traversal around given nodes
`create_nodes`	Create or merge-update nodes of any type
`create_relations`	Create edges with context, weight, and confidence
`add_sources`	Record provenance (Source nodes + DERIVED_FROM links)
`get_temporal_sequence`	Follow chronological chains (NEXT, BEFORE, CAUSES)
`create_reasoning_chain`	Build structured multi-step reasoning
`get_reasoning_chain`	Retrieve reasoning chains by name or topic

Every tool declares an input schema (validated), an output schema, and behavior annotations (read-only vs. write), and returns machine-readable structuredContent alongside a JSON text block. Writes merge: re-creating an existing node or edge updates the fields you provide and preserves the ones you don’t.

The server also exposes MCP prompts — a system prompt with the overall tool-usage workflow, plus per-tool usage guidance under each tool’s name.

Node Types (15)

Entity, Event, Concept, Attribute, Proposition, Emotion, Agent, ScientificInsight, Law, Location, Thought, ReasoningChain, ReasoningStep, Source, EmotionalEvent

Storage

SQLite with WAL mode for concurrent reads
FTS5 virtual table for full-text search with BM25 ranking
Recursive CTEs for graph traversal
Single nodes table for all types, edges table with UNIQUE constraint, plus aliases and observations tables
Complex objects stored as JSON in a properties column

Installation

Prerequisites

Node.js v18+
A C/C++ toolchain for compiling the better-sqlite3 native module:
- Mac: Xcode Command Line Tools (xcode-select --install)
- Linux: build-essential (apt install build-essential)
- Windows: Visual Studio Build Tools

From npm

npx mcp-engram

From source

npm install
npx nx build mcp-engram
node dist/servers/mcp-engram/main.js

Environment Variables

Variable	Required	Default	Description
`MCP_ENGRAM_DB_PATH`	No	`~/.mcp-engram/knowledge.db`	Path to the SQLite database file

Integrating with Claude Desktop

在 GitHub 查看完整 README

安装

npx mcp-engram

新手？先看 mcpskill.org 教程 →

配置

{
  "mcpServers": {
    "mcp-engram": {
      "command": "npx",
      "args": ["mcp-engram"]
    }
  }
}