ja-local-kb
Obsidian is the single source of truth for the knowledge base. An Obsidian plugin provides source registration, state control, watcher-based serial chunking and vectorization, a status page, file tree/graph syncing, and connection control for Claude and Codex agents.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ja-local-kbrecall what my notes say about the Qwen3-Embedding-8B benchmark results"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
JA Local KB
本目录是跨平台产品实现,和上一级 40_code 中的 Windows/Ollama 历史原型分开。
当前状态
0.1.0 冻结了已经完成 Windows 实机验收的 dev.12:高召回、自动增量、
Obsidian 控制面、相邻版本回滚与设备来源保护均已进入首个公开稳定 Release。
支持范围必须分开理解:Windows 已验收;macOS 脚本已提供但仍等待真实 Mac
端到端验收;Claude 真实客户端也尚未验收。
已实现显式来源注册、结构化 Chunk、差异 Embedding、SQLite 状态、 LanceDB 全文/向量/混合检索和失效拒答。
已实现且只公开三个本地 STDIO MCP 工具。
recall是默认 Agent 模式:8B Embedding 生成最多 80 条可追溯证据, 直接交给 Claude/Codex 二次筛选、分析和回答,不调用 Reranker。quality保留为可选紧凑模式:独立 API Reranker 将同一候选池压缩到 24-40 条;它不是默认主链路。已构建 Obsidian 来源与状态控制插件:右键显式入库后释放弹窗, watcher 后台串行切片和向量化,状态页显示进度,文件树与原生图谱同步标记; 无变化轮询不再重建侧栏,真实更新会保留滚动与全部折叠状态。
已加入默认收起的 Codex、Claude 紧凑连接入口:复制完整提示词后由目标 Agent 检查环境、 备份并合并 MCP 配置,再以真实工具发现和状态/检索/证据回读验证连接。
“检索诊断”只展示本次候选证据,不宣称找全;“召回质量报告”只对冻结 题目和人工标注证据给出正式结论,并在语料、索引、模型或配置变化后标旧。
已提供 Windows/macOS 的预检、安装、诊断、更新和回滚脚本;版本快照不保存
sources.json,旧快照恢复也不能覆盖设备当前来源。已使用 42 份冻结基准来源、783 个 Chunk 和 12 道冻结交叉项目题完成 4B/8B 同条件 API 验收。
历史冻结基准的主链路为
Qwen/Qwen3-Embedding-8B + recall Top80:证据覆盖47/47(100%)、项目覆盖27/27(100%)、可追溯100%。上述结果对应 42 份来源和索引版本 131;当前来源范围变化后,系统会将 它标为历史/过期,不能作为当前语料已经完整召回的证明。
本次正式单轮复杂题延迟中位数
3.112 秒、P9535.914 秒,召回质量 通过,但 API 长尾速度门槛未通过,必须与质量指标分开呈现。8B Reranker Top40 为
42/47(89.36%),说明紧凑压缩会丢证据; 仅在上下文预算不足时按需启用。真实 Obsidian、Codex MCP、来源保护与 Windows 相邻版本回滚已经通过; macOS 与 Claude 客户端仍需分别完成真实端到端验证。
Related MCP server: enquire-mcp
产品边界
Obsidian Markdown 是唯一真源。
每台设备独立保存来源清单、SQLite 状态与 LanceDB。
建库与查询统一使用 API Embedding。
v1 对 Agent 只公开本地 STDIO MCP。
Obsidian 插件是来源、状态、操作和连接控制面,不提供聊天回答。
首版通过公开 GitHub 稳定 Release 与本地 Agent 辅助部署;正式设备只安装 已审核 tag/Release,不自动追踪
main。
开发与发布只允许使用无真实业务数据的测试夹具。Vault、LanceDB、日志、 API Key、来源注册表和机器配置不得进入 GitHub 或 Release。
获取与授权边界
代码仓库:https://github.com/jefferyadventure519-byte/ja-local-kb
稳定版本:https://github.com/jefferyadventure519-byte/ja-local-kb/releases
本仓库公开是为了简化设备安装和更新,不等于开源;当前没有附加开源许可证。 复制、修改或二次分发前,应先取得仓库所有者许可。
目录入口
docs/architecture.mddocs/state-machine.mddocs/mcp-contract.mdcontracts/source-registry.schema.jsonINSTALL_AGENT.mdUPDATE_AGENT.md
本地开发验证
Python 3.11+、uv 和 Node.js 20+ 可用后:
$runtimeRoot = 'D:\JA-Local-KB'
$vaultRoot = 'D:\path\to\your-vault'
$env:UV_CACHE_DIR = Join-Path $runtimeRoot 'runtime\uv-cache'
uv sync --all-extras --no-editable
uv run --no-sync ruff check .
pwsh -File scripts\test-windows.ps1 `
-RuntimeRoot $runtimeRoot `
-VaultRoot $vaultRoot
npm.cmd --prefix obsidian-plugin run build
uv buildWindows 中文路径下使用 --no-editable,避免某些 .pth 加载器按系统编码读取
UTF-8 路径。修改 Python 源码后需重新执行 uv sync ... --reinstall-package ja-local-kb 再测试。Windows 测试必须通过 scripts/test-windows.ps1 运行;
脚本会先同步当前源码,把 pytest 临时目录固定到所选运行根目录的
tmp\pytest\,复用 Vault 外的运行时 uv 缓存,并拒绝调用方覆盖
--basetemp。不得在任何 Obsidian Vault 内创建 pytest 缓存、临时目录或
隔离目录。
官方兼容依据
OpenAI Embeddings API: https://developers.openai.com/api/reference/resources/embeddings/methods/create
Qwen3 Embedding:https://qwenlm.github.io/blog/qwen3-embedding/
SiliconFlow Embeddings API: https://docs.siliconflow.cn/en/api-reference/embeddings/create-embeddings
SiliconFlow Rerank API: https://docs.siliconflow.cn/cn/api-reference/rerank/create-rerank
Qwen3-Reranker-8B: https://huggingface.co/Qwen/Qwen3-Reranker-8B
MCP Python SDK:https://github.com/modelcontextprotocol/python-sdk
Obsidian 插件开发:https://docs.obsidian.md/Plugins/Getting%20started/Build%20a%20plugin
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- Alicense-qualityDmaintenanceLocal MCP server for indexing personal knowledge into SQLite with hybrid search, chunk-level citations, memory tools, and agent orchestration.Last updated4MIT
- AlicenseAqualityBmaintenanceMCP server that exposes a local Obsidian vault as persistent, searchable memory for AI agents, with hybrid retrieval, reranker, and PDF support.Last updated342,37128MIT
- Alicense-qualityAmaintenanceAn MCP server that provides semantic search and vault operations for Obsidian notes, enabling agents to recall and manage notes with provenance and low token usage.Last updated276ISC
- Alicense-qualityBmaintenanceA personal memory MCP server for Obsidian/markdown vaults that provides hybrid retrieval (semantic, keyword, tag, time-aware co-occurrence) and is shared across multiple AI clients.Last updated324MIT
Related MCP Connectors
Local-first RAG engine with MCP server for AI agent integration.
Serve a folder of Markdown notes as an MCP server: hybrid search, reading, and sourced answers.
Token-efficient MCP memory for Markdown vaults. Tiered search, GraphRAG, AI memories.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/jefferyadventure519-byte/ja-local-kb'
If you have feedback or need assistance with the MCP directory API, please join our Discord server