Ruminer Browser Agent
Syncs AI conversations from multiple platforms to local Markdown files for your second brain, and provides browser automation tools via MCP for agentic workflows.
README
Ruminer Browser Agent
Sync your AI conversations from 5+ platforms into your second brain — quick search, agent-ready, yours forever.
- Continuously export conversations to local Markdown files across AI chat platforms: ChatGPT, Gemini, Claude, DeepSeek, and Grok.
- Your user credentials on these platforms stay secure in your own browser, never uploaded to cloud.
- Freely choose your AI agent backend: OpenClaw, Claude Code, or Codex.
- Make your agent understand you deeply via RAG from your local Markdown conversation archive.
- Seamlessly integrated into your Chrome browser with beautiful UI and powerful automation.
https://github.com/user-attachments/assets/fb49d045-3d43-49b6-b51a-3ee8ff9ed508



System Architecture
flowchart TB
subgraph MCP_Clients[MCP Clients]
Codex[Codex CLI]
ClaudeCode[Claude Code]
OpenClawPlugin[OpenClaw]
end
subgraph OpenClaw[OpenClaw Gateway]
Gateway[ws://127.0.0.1:18789]
end
subgraph Ruminer[Ruminer Stack]
NativeServer[Messaging Server]
McpServer[MCP Server<br/>http://127.0.0.1:12306/mcp]
Extension[Chrome Extension<br/>Service Worker]
Sidepanel[Side Panel UI]
Workflows[Ingestion Workflows]
end
subgraph Chrome[Chrome Browser Session]
APIs[Chrome APIs<br/>tabs, read page, scrolling, click]
end
subgraph Memory[Local Memory]
LocalMD[Markdown Files<br/>~/your-memory-dir/]
end
subgraph AIPlatforms[AI Chat Platforms]
ChatPlatforms[ChatGPT, Gemini, Claude, DeepSeek, Grok]
end
MCP_Clients <-->|"Streamable HTTP"| McpServer
APIs <-->|"Browser Tools"| McpServer
LocalMD -->|"RAG Tools"| McpServer
NativeServer <-->|"Native Messaging"| Extension
Extension <--> APIs
Sidepanel <-->|"Chat"| NativeServer
NativeServer <-->|"WebSocket"| Gateway
Workflows -->|"Write Markdown"| LocalMD
LocalMD -->|"Memory Search"| Sidepanel
AIPlatforms -->|"Conversation Data"| Workflows
The Ruminer UI has three major pillars in one Chrome extension:
- Chat tab: communicate with your CLI agents via native server, providing MCP tools for browser operations and local RAG.
- You can toggle tool groups (Memory / Observe / Navigate / Interact / Execute / Workflow) in the message input box to control which ones the agent can use.
- Memory tab: browse/search/manage your local memory directory.
- Conversations are organized by AI chat platform and stored as plain Markdown files you own.
- Workflows tab: create, edit, and schedule automation workflows to export conversations to your local Markdown directory or accomplish other tasks in browser.
- Coming soon: agent-driven workflow development by autonomously interacting with the browser and editing the workflow graph!
Glossary
- OpenClaw Gateway: local control plane for chat + tool runtime (Ruminer sidepanel chat talks to it).
- MCP: Model Context Protocol; here it’s the standard interface your clients use to call browser tools.
- Local Markdown memory: plain
.mdfiles written to a directory you configure — compatible with Obsidian, Logseq, and any agent that can read files (OpenClaw, Hermes, GBrain, Claude Code, Codex).
Getting Started (local dev)
Prerequisites
- Node.js
>= 22.5.0 pnpm(seepackage.json)- Chrome/Chromium (MV3 + sidepanel enabled)
- Optional but recommended:
openclawCLI (for sidepanel chat + plugin routing)- A local Markdown directory path (for memory storage + RAG ingestion)
1) Quick setup
From the repo root, run:
pnpm install
pnpm -C app/chrome-extension build
Then open the extension Welcome page (it opens automatically on install) and run the one‑shot installer to set up Native Messaging + MCP clients:
curl -fsSL https://raw.githubusercontent.com/casterkay/ruminer-browser-agent/refs/heads/main/scripts/setup.sh | \
bash -s -- --extension-id <your-extension-id>
This installer (best-effort):
- Installs
chrome-mcp-serverand registers the Native Messaging host (allowlisted to your extension ID) - Adds Ruminer MCP endpoint to Claude Code and Codex (
ruminer-chrome) - Installs + enables the OpenClaw plugin (
openclaw-mcp-plugin) and points it to Ruminer MCP URL
2) Load the extension
Chrome does not allow “Load unpacked” via script.
- Open
chrome://extensions - Enable Developer mode
- Click Load unpacked
- Select:
app/chrome-extension/.output/chrome-mv3
3) Configure the extension
In the Settings tab in Ruminer side panel:
- OpenClaw Gateway
- WS URL:
ws://127.0.0.1:18789 - Token: your Gateway token
- WS URL:
- Memory Directory
- Local path to your Markdown memory folder (e.g.
~/Documents/my-second-brain)
- Local path to your Markdown memory folder (e.g.
Hosted auth and billing URL
Workflow unlock now relies on the hosted Ruminer web app in landing-page for Better Auth + Stripe.
The extension opens that app for sign-in, checkout, and account management, then receives a browser-bound
access snapshot back through the background service worker.
- Set
WXT_PUBLIC_RUMINER_WEB_URLto your hosted web app origin when building or running the extension. VITE_RUMINER_WEB_URLis also accepted as a fallback alias if you already use that naming in local env files.- If neither env var is set, the extension defaults to
http://localhost:3000.
Local dev example:
cd landing-page
pnpm install
pnpm dev
cd ../app/chrome-extension
WXT_PUBLIC_RUMINER_WEB_URL=http://localhost:3000 pnpm dev
Verify It Works
- MCP tool check:
- Ask the agent to call a tool (e.g. "List the current tab titles in my browser") in CLI.
- Side panel chat:
- Open Ruminer side panel → Chat → send a message → see tool calls render inline
- Memory suggestions (requires memory directory configured):
- Type ≥ 3 characters in message input box → see debounced suggestions from your local Markdown files
- Workflows (requires memory directory configured):
- Open Workflows tab → run a built‑in workflow → re-run should not duplicate (ledger + stable IDs)
Developer Notes
This is a pnpm workspace with key packages:
app/chrome-extension: MV3 extension (Vue 3 + WXT + Tailwind)app/native-server: Fastify server + Native Messaging host + MCP transportpackages/shared: shared types + tool schemas
License
AGPL-3.0 (see LICENSE).
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。