GitHub 宝藏挖掘 - 2026-08-16¶
Top 5¶
1. DeusData/codebase-memory-mcp (39,053⭐) [🔴]¶
- Score: 87/100 | Lang: C | License: MIT
- High-performance code intelligence MCP server. Indexes codebases into a persistent knowledge graph — average repo in milliseconds. 158 languages, sub-ms queries, 99% fewer tokens. Single static binary
- URL: https://github.com/DeusData/codebase-memory-mcp
2. HKUDS/nanobot (47,047⭐) [🔴]¶
- Score: 81/100 | Lang: Python | License: MIT
- Ultra-lightweight, open-source, self-hosted personal AI agent framework in Python with WebUI, tools, memory, MCP, multi-agent workflows, automation, and chat apps
- URL: https://github.com/HKUDS/nanobot
3. tirth8205/code-review-graph (30,287⭐) [🔴]¶
- Score: 72/100 | Lang: Python | License: MIT
- Local-first code intelligence graph for MCP and CLI. Builds a persistent map of your codebase so AI coding tools read only what matters, with benchmarked context reductions on reviews and large-repo w
- URL: https://github.com/tirth8205/code-review-graph
4. ChromeDevTools/chrome-devtools-mcp (49,233⭐) [🟡]¶
- Score: 69/100 | Lang: TypeScript | License: Apache-2.0
- Chrome DevTools for coding agents
- URL: https://github.com/ChromeDevTools/chrome-devtools-mcp
5. langgenius/dify (152,564⭐) [🟡]¶
- Score: 62/100 | Lang: TypeScript | License: NOASSERTION
- Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without reb
- URL: https://github.com/langgenius/dify
MCP 生态发现¶
- DeusData/codebase-memory-mcp (87/100): High-performance code intelligence MCP server. Indexes codebases into a persistent knowledge graph — average repo in milliseconds. 158 languages, sub-
- HKUDS/nanobot (81/100): Ultra-lightweight, open-source, self-hosted personal AI agent framework in Python with WebUI, tools, memory, MCP, multi-agent workflows, automation, a
- tirth8205/code-review-graph (72/100): Local-first code intelligence graph for MCP and CLI. Builds a persistent map of your codebase so AI coding tools read only what matters, with benchmar
洞察¶
- MCP 持续成为 Agent 标准接口
- 本地优先是主流趋势
- 记忆系统是核心突破点
自动生成 2026-08-16 13:50
🗺️ 属于 [[MOC-Research]] · [[Home|🏠 Home]]
周报处理评估（2026-08-17 k 追加）¶
用户确认处理。5 项目逐项评估结论：
| 项目 | 结论 | 依据 |
|---|---|---|
| mattpocock/skills | ✅ 已吸收（7-22 起） | mattpocock-methodology.md + engineering-workflow + grill-with-docs 全是它的适配 |
| stablyai/orca | ❌ 不装 | 已有 Hermes+dsh 桌面端；用户偏好轻量自建 |
| firecrawl | ❌ 不需要 | Hermes 已有 web_extract/web_search |
| HKUDS/DeepTutor | ✅ 已评估 | github-projects-note.md 架构吸收笔记 |
| 3b1b/manim | ⏸️ 暂缓 | LaTeX 环境成本高，B站教学暂未启动 |
唯一增量落地：wayfinder（多会话大任务规划）模式 → 已补进 engineering-workflow 技能 §⑥.8。 触发条件：任务大到一个 agent 会话装不下且路线不清；地图=Obsidian 笔记，决策 ticket=Hermes todo。