Recalla
Enables AI agents to read and search your Obsidian vault through a local MCP server, keeping everything private and local.
README
Recalla for Obsidian
Make your Obsidian vault a fast, private, agent-readable brain. Recalla indexes your
notes into agent-readable cards, serves a local HTTP search API on 127.0.0.1:8090,
and exposes an MCP server so AI agents can read your vault. Everything runs locally,
and you never have to touch a terminal.
The plugin runs a small standalone Recalla engine in the background and adds an in-Obsidian chat panel and search panel plus one-click setup commands. Nothing to install by hand: the plugin downloads the engine for your operating system on demand.
What it does
- Downloads and manages the Recalla engine automatically (no Python, no terminal).
- Indexes your vault into agent-readable cards.
- Runs a local HTTP API (
127.0.0.1:8090by default) for fast search. - Adds a right-sidebar chat panel that answers from your vault with clickable source links.
- Provides an in-Obsidian search panel backed by that API.
- Copies a ready-to-paste MCP config so AI agents can connect to your vault.
- Generates
AGENTS.mdandllms.txtfor agent tooling.
Requirements
- Obsidian 1.4.0 or newer (desktop only, since the plugin runs a local engine).
That is it. The first time you need the engine, run Recalla: Make my vault agent-readable or Recalla: Download or update the engine and the plugin fetches the correct standalone binary from the latest GitHub release, caches it under the plugin folder, and uses it for everything.
Advanced users can point the plugin at their own engine build via the
Custom engine path setting. Resolution order is: custom path, then the
downloaded engine, then recalla on your PATH.
Answering questions
The chat panel runs recalla ask locally. To get answers you need a language
model:
- Set
OPENAI_API_KEYin your environment for OpenAI, or - Set
ANTHROPIC_API_KEYfor Claude, or - Run a local model with Ollama, which needs no key.
Restart Obsidian after setting an environment variable so it is picked up.
Install (manual)
- Build or download
main.jsand grabmanifest.jsonfrom this repo. - Create the folder
<your-vault>/.obsidian/plugins/recalla/. - Copy
main.jsandmanifest.json(andstyles.cssif present) into that folder. - In Obsidian, open Settings, go to Community plugins, and enable "Recalla".
Community-plugin store submission is planned, after which you will be able to install it directly from inside Obsidian.
Build from source
npm install
npm run build
This produces main.js in the repo root.
Usage
Click the chat ribbon icon, or open the command palette and run any of these:
Recalla: Make my vault agent-readabledownloads the engine if needed, indexes the vault, and starts the local server, all in one click.Recalla: Download or update the enginefetches the latest engine binary.Recalla: Open chatopens the right-sidebar chat panel.Recalla: Reindex vaultrebuilds the agent index.Recalla: Start local serverstarts the local HTTP API and MCP server.Recalla: Stop local serverstops it.Recalla: Search vault (agent index)opens the in-Obsidian search panel.Recalla: Copy MCP config to clipboardcopies an MCP server config for your agent.Recalla: Generate AGENTS.md and llms.txtwrites agent-tooling files into the vault.
A status bar item shows the current state: Recalla: off, Recalla: running :8090,
or Recalla: indexing....
Settings
- Recalla engine: shows engine status and a button to download or update it.
- Custom engine path: optional override. Leave empty to use the managed engine.
- Server port: local HTTP API port (default
8090, bound to127.0.0.1). - Auto-start server on load: start the server when Obsidian opens.
- Auto-reindex on save: reindex automatically after notes change (debounced).
License
Apache-2.0
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。
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。
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。
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。