Hermes Memory MCP
A local enhanced memory system with self-evolution capabilities, enabling automatic skill creation, periodic memory maintenance, and trajectory training for reinforcement learning.
README
🧠 Hermes Memory MCP
本地增强型记忆系统 + 自我进化能力 · v2.3
基于 Model Context Protocol (MCP) 的本地记忆系统,5 个版本演进到 v2.3 自我进化阶段:从被动存储 → 主动学习 → 智能检索 → 自我进化。
✨ v2.3 三大自进化能力
| 能力 | 说明 | MCP Tools |
|---|---|---|
| Skill 自动创建 | 从对话上下文抽取 SKILL.md 草稿 → 审核 → 一键部署到 Trae | auto_create_skill / approve_draft_skill / list_draft_skills |
| 周期性 nudge 调度 | 6 个内置后台任务自动维护记忆库(清理/压缩/反思/晋升/健康检查) | trigger_nudge / list_nudges |
| Trajectory 训练管道 | 自动捕获所有 MCP Tool 调用 → 导出 JSONL → 用于 RLHF/DPO 微调 | export_trajectory / mark_trajectory_reward / get_trajectory_stats |
📊 项目指标
| 维度 | 数据 |
|---|---|
| 端到端测试 | 51 / 51 通过(v2.1=23 + v2.2=11 + v2.3=17) |
| MCP Tool 总数 | 31(v2.1=15 + v2.2=7 + v2.3=9) |
| Skill 数 | 5(memory / evolution / knowledge / reflection / auto-orchestrator) |
| 6 个 nudge | clean_expired / compress_long_session / reflect_unreflected / cold_memory_warmup / cross_project_promote / health_check |
| 数据 Schema | 20 张表(兼容升级 v2.1 老数据) |
🏗️ 架构
hermes-mcp/
├── hermes_memory_mcp/ # MCP Server 核心
│ ├── server.py # 31 个 @mcp.tool() 入口
│ ├── core/ # MemoryManager(5 版本核心)
│ ├── storage/ # SQLite + ChromaDB
│ ├── orchestrator/ # v2.2 + v2.3 新增
│ │ ├── multisignal.py # v2.2 多信号融合
│ │ ├── system_prompt.py # v2.2 三层 Prompt
│ │ ├── pre_compress.py # v2.2 压缩前捕获
│ │ ├── skill_factory.py # v2.3 Skill 自动创建
│ │ ├── scheduler.py # v2.3 nudge 调度
│ │ ├── trajectory.py # v2.3 Trajectory 训练管道
│ │ └── nudges/ # v2.3 6 个后台任务
│ └── utils/ # audit / sanitizer / embedder
├── skills/ # 5 个 Trae Skill
│ └── hermes-{memory,evolution,knowledge,reflection,auto-orchestrator}/
├── tests/ # e2e + 报告 + 复盘 + 打包
│ ├── test_e2e.py # v2.1 基础(23 用例)
│ ├── test_v22.py # v2.2 多信号 + prompt(11 用例)
│ ├── test_v23.py # v2.3 自进化(17 用例)
│ ├── deployment_report.md # 部署报告(v2.1 → v2.2 → v2.3)
│ ├── RETROSPECTIVE_v23.md # v2.3 完整复盘
│ └── build_v23_bundle.py # 打包脚本
├── .trae/
│ ├── mcp.json # Trae MCP 配置
│ └── specs/ # 7 个 spec 文档
│ ├── debug-and-deploy-hermes-mcp/
│ ├── enhance-hermes-memory-mcp/
│ ├── reference-hermes-agent-online/
│ ├── trae-auto-memory/
│ ├── v22-evolution/
│ ├── v23-self-evolution/
│ ├── v23-retrospective-bundle/
│ └── github-publish-v23/
└── hermes_memory_mcp_server.py # 入口脚本
🚀 快速开始
# 1. 安装依赖
pip install mcp chromadb sentence-transformers networkx jieba pyyaml aiofiles
# 2. 配置 Trae MCP(.trae/mcp.json 已就绪)
# command: C:/Python314/python.exe
# args: e:/hermes-project/hermes-mcp/hermes_memory_mcp_server.py
# 3. 启动(默认离线 Mock Embedder;设 HERMES_USE_REAL_EMBED=1 用真实模型)
python hermes_memory_mcp_server.py
# 4. 跑端到端测试
python tests/test_e2e.py # 23/23
python tests/test_v22.py # 11/11
python tests/test_v23.py # 17/17
# 5. 打包 v23 交付物
python tests/build_v23_bundle.py
🧬 5 大版本演进
| 版本 | 阶段 | 关键能力 | 状态 |
|---|---|---|---|
| v2.0 | 基础 | retain/recall/reflect/evolve | ✅ 17 e2e |
| v2.1 | 自动化 | auto_retain/startup_recall/summarize_session | ✅ 23 e2e |
| v2.2 | 智能化 | 多信号检索 / 三层 prompt / on_pre_compress / Linked / self-healing | ✅ 34 e2e |
| v2.3 | 自我进化 | Skill 自动创建 / 6 nudge / Trajectory 训练管道 | ✅ 51 e2e |
📐 业界借鉴
- NousResearch/hermes-agent → 三层 System Prompt(v2.2)
- Letta → on_pre_compress + 反思代理独立化(v2.2 / v2.3)
- Mem0 → 多信号融合 + Linked Memory(v2.2)
📝 复盘与部署
- 完整复盘:tests/RETROSPECTIVE_v23.md
- 部署报告:tests/deployment_report.md
- 打包产物:
tests/v23_bundle_*.zip(48.5KB / 16 文件)
📜 License
Private · 仅供 owner=wcy88 内部使用
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.