AgentClip MCP Server
Enables AI agents to store, search (keyword and semantic), retrieve, update, and delete personal browser clips from a Cloudflare-backed store, authenticated via per-user token.
README
AgentClip
Clip selected browser text — with a confirm/edit step — into a personal store on Cloudflare, then retrieve it from AI agents (MCP) and tools over a per-user-token-authenticated REST API.
- agent — AI clients reach your clips over MCP / REST with a per-user token.
- clip — a Chrome extension (Manifest V3) adds a "Save to AgentClip" context menu; you edit the text before it's saved.
Production runs as a single Cloudflare Worker at https://agentclip.0xkaz.com serving
the API, the MCP server, Google OAuth, and the dashboard from one origin.
- Live app: https://agentclip.0xkaz.com
- Source: https://github.com/0xkaz/agentclip
- 日本語版 README: README.ja.md
Features
- Clip & edit — right-click any selection → edit content/title/tags/source → save.
- Keyword search — substring match over content/title/tags (works for Japanese too).
- Semantic search — meaning-based search via Workers AI embeddings + Vectorize.
- Per-clip encryption — mark a clip to encrypt it at rest; encrypted clips are excluded from search (and shown masked, with a reveal toggle).
- Share links — publish a single clip at a public read-only
/s/<slug>URL; revoke anytime (encrypted clips can't be shared). - MCP + REST — AI agents and tools read/write your clips with a per-user token.
- Google sign-in for the dashboard; opaque per-user Bearer tokens for machines.
Use the hosted app
The fastest path — no local setup:
- Open https://agentclip.0xkaz.com and Sign in with Google.
- Go to API Tokens → create a token (
ac_live_…, shown once — copy it). - Open Setup & Usage in the app — it shows your exact MCP config and
curlexamples, plus the Chrome-extension steps below.
The dashboard has three pages:
| Page | What it does |
|---|---|
| Snippets | Browse, add/edit/delete, keyword & semantic search your clips |
| API Tokens | Create / revoke per-user Bearer tokens |
| Setup & Usage | Copy-paste MCP config, REST curl, and extension setup |
Monorepo layout
apps/api Cloudflare Worker — REST + MCP + Google OAuth, D1, serves dashboard
apps/web React + Vite dashboard — landing page + Snippets / Tokens / Setup
apps/extension Chrome MV3 extension — capture + confirm/edit
packages/shared shared TypeScript types
The rest of this README covers running and deploying it yourself.
Requirements: Node 20+, pnpm 9+ (npm i -g pnpm), a Cloudflare account, and a
Google OAuth client.
1. Clone & install
git clone https://github.com/0xkaz/agentclip.git
cd agentclip
pnpm install # or: make install
2. Configure Cloudflare D1
pnpm --filter @agentclip/api exec wrangler d1 create agentclip
Paste the returned database_id into both [[d1_databases]] blocks in
apps/api/wrangler.toml (the top-level one and the [env.production] one).
make db-init # apply schema.sql to the LOCAL D1
# later, for production:
pnpm --filter @agentclip/api db:init:remote
3. Configure Google OAuth
In Google Cloud Console → Credentials → create an OAuth client ID (Web application). Add authorized redirect URIs:
- Local:
http://localhost:8787/auth/google/callback - Production:
https://agentclip.0xkaz.com/auth/google/callback
Then create local secrets:
cp apps/api/.dev.vars.example apps/api/.dev.vars
# edit apps/api/.dev.vars:
# GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET, SESSION_SECRET (any random 32+ byte string)
4. Run locally
make dev # API + MCP on http://localhost:8787
pnpm --filter @agentclip/web dev # dashboard on http://localhost:5173
Open http://localhost:5173, sign in with Google, and create an API token.
5. Install the Chrome extension
pnpm --filter @agentclip/extension build # outputs apps/extension/dist
- Open
chrome://extensions→ enable Developer mode. - Load unpacked → select
apps/extension/dist. - Click the AgentClip icon → set API base URL and paste the token from the dashboard.
- Local:
http://localhost:8787 - Production:
https://agentclip.0xkaz.com
- Local:
- Select text on any page → right-click → Save to AgentClip → edit → Save.
6. Deploy to production (agentclip.0xkaz.com)
# one-time: set production secrets
pnpm --filter @agentclip/api exec wrangler secret put GOOGLE_CLIENT_ID --env production
pnpm --filter @agentclip/api exec wrangler secret put GOOGLE_CLIENT_SECRET --env production
pnpm --filter @agentclip/api exec wrangler secret put SESSION_SECRET --env production
pnpm --filter @agentclip/api db:init:remote # schema on remote D1
make deploy # builds the dashboard, then deploys the Worker (--env production)
The custom domain route (
agentclip.0xkaz.com) requires the Cloudflare API token to have Zone:Edit on0xkaz.comin addition to Workers/D1 edit. Without it, deploy the Worker and bind the domain from the dashboard, or use the*.workers.devURL.
Using the REST API
# save (add "encrypted": true to encrypt at rest + exclude from search)
curl -X POST https://agentclip.0xkaz.com/api/snippets \
-H "Authorization: Bearer ac_live_..." \
-H "Content-Type: application/json" \
-d '{"content":"hello","tags":"demo"}'
# keyword search (substring; works for Japanese)
curl "https://agentclip.0xkaz.com/api/snippets?q=hello" -H "Authorization: Bearer ac_live_..."
# semantic search (meaning-based)
curl "https://agentclip.0xkaz.com/api/snippets?q=greeting&mode=semantic" -H "Authorization: Bearer ac_live_..."
# update / delete
curl -X PATCH https://agentclip.0xkaz.com/api/snippets/1 -H "Authorization: Bearer ac_live_..." -H "Content-Type: application/json" -d '{"content":"updated"}'
curl -X DELETE https://agentclip.0xkaz.com/api/snippets/1 -H "Authorization: Bearer ac_live_..."
Using MCP
Point an MCP client at https://agentclip.0xkaz.com/mcp with header
Authorization: Bearer ac_live_.... Tools: store_snippet, search_snippets,
semantic_search, get_snippet, list_recent, update_snippet, delete_snippet.
Semantic search indexes new clips automatically; to index clips made earlier, open the Snippets page, switch to Semantic, and click Rebuild index (or
POST /api/my/reindex). Vectorize is eventually consistent (new clips become searchable within ~1 minute).
See docs/architecture.md and docs/agent-workflow.md for details.
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。