project-brain-mcp
Engineering memory for Claude Code — prevents re-investigating solved problems and repeating rejected architectural decisions across sessions and projects.
README
Project Brain MCP
Engineering memory for Claude Code — prevents re-investigating solved problems and repeating rejected architectural decisions across sessions and projects.
What it does
- Prevent re-investigating: Answered questions are recorded as findings. Future sessions skip re-investigation.
- Prevent repeating rejected architectures:
validate_planchecks your proposal against past decisions and mistakes before you proceed. - Cross-project learning: Decisions from other projects appear as soft references, not hard blocks.
- Persistent memory: Survives across sessions, stored in
~/.project-brain/memory.json.
Installation
pip install git+https://github.com/pym2282/project-brain-mcp
claude mcp add project-brain project-brain-mcp --scope user
Restart Claude Code. The MCP server is now active in all your projects.
Uninstall
pip uninstall project-brain-mcp
claude mcp remove project-brain --scope user
Memory location
Memory is stored at ~/.project-brain/memory.json — shared across all projects, never committed to any repo.
To use a custom path:
export PROJECT_BRAIN_MEMORY_PATH=/path/to/memory.json
Tools
| Tool | When to call |
|---|---|
get_context() |
Session start — returns slim index |
get_context(focus=[...]) |
When index shows relevant entries — returns full content |
validate_plan(plan, current_project) |
Before any architectural proposal |
add_decision(title, reason, tags, project) |
When a decision is confirmed |
add_finding(question, conclusion, tags, project) |
When an investigation question is answered |
add_mistake(description, lesson, tags, project) |
When a wrong approach is identified |
update_entry(id, updates) |
To correct an existing entry |
delete_entry(id) |
To remove an outdated entry |
update_open_questions(questions) |
To track unresolved questions |
How validate_plan works
validate_plan("use sqlite for storage", current_project="my-app")
→ conflicts: [entries from "my-app" that match — must resolve]
→ references: [entries from other projects — consider as context]
Same project matches are hard conflicts. Other project matches are soft references.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。