rainkeeper
Exposes the Raindrop.io API as Claude tools for managing bookmarks, collections, tags, and bulk operations, with proper collection move support.
README
· · · · · · · · · · · · · · · ·
· · · · · · · · · · · · · ·
____ _ ___ _ _ _ _______ _____ _____ ____ _____ ____
| _ \ / \ |_ _| \ | | |/ / ____| ____| ___| _ \| ____| __ \
| |_) | / _ \ | || \| | ' /| _| | _| | |_ | |_) | _| | /
| _ < / ___ \ | || |\ | . \| |___| |___| _| | __/| |___| _ \
|_| \_\/_/ \_\___|_| \_|_|\_\_____|_____|_| |_| |_____|_| \_\
· · · · · · · · · · · · · ·
· · · · · · · · · · · · · · · ·
<div align="center">
The Raindrop.io MCP Server that works on free accounts — and actually moves bookmarks.
</div>
What is Raindrop.io?
Raindrop.io is the bookmark manager that makes the web collectible. Save anything — articles, videos, PDFs, images, links — organise it into collections and sub-collections, tag it, annotate it, and surface it exactly when you need it. It runs on every platform, syncs everywhere, and has a free tier that covers everything most people ever need.
If you've ever lost a link, abandoned browser bookmarks, or wished your read-later queue could talk back — Raindrop is the answer.
Rainkeeper connects Raindrop to Claude, giving you natural-language control over your entire library.
Why Rainkeeper?
Two problems with the existing MCP options:
The official Raindrop.io MCP server requires a Pro subscription (~$3/month). If you're on the free tier — which covers the vast majority of Raindrop's feature set — you cannot use it.
The community alternative (adeze/raindrop-mcp) has a silent critical bug: collection moves do nothing. Ask Claude to file a bookmark into a collection, and the API call returns 200 OK — while the bookmark sits exactly where it was. The root cause is a malformed request body that the Raindrop API silently ignores. The same bug breaks all bulk move operations.
Rainkeeper fixes both.
| Rainkeeper | Official MCP | Community MCP | |
|---|---|---|---|
| Free account | ✅ | ❌ Pro only | ✅ |
| Collection moves work | ✅ | ✅ | ❌ Silent fail |
| Sub-collection support | ✅ | — | ❌ |
| Bulk operations | ✅ | — | ✅ (broken) |
| Self-hostable | ✅ | ✅ | ✅ |
| Open source (MIT) | ✅ | — | ✅ |
The Fix
The Raindrop.io REST API uses a nested object for all collection references. Sending a flat integer is accepted without error — and silently ignored.
// ✅ Correct — what Rainkeeper sends
PUT /raindrop/123
{"collection": {"$id": 456}}
// ❌ Silently ignored — what the community MCP sends
PUT /raindrop/123
{"collectionId": 456}
The same applies to bulk operations:
// ✅ Correct
PUT /raindrops/0
{"ids": [111, 222], "collection": {"$id": 456}}
// ❌ Returns 404 or silently fails
PUT /raindrops/0
{"ids": [111, 222], "collectionId": 456}
RaindropClient enforces the correct format at the HTTP layer. Tools never pass raw integer IDs to the API. One rule, one place, always correct.
Quick Start
1. Get your token
Raindrop.io → Settings → Integrations → For Developers → create an app → copy the Test token. Free accounts included. Takes 30 seconds.
2. Install
git clone https://github.com/simonives/rainkeeper.git
cd rainkeeper
pip install -e . # or: uv pip install -e .
3. Configure
cp .env.example .env
# Open .env — set RAINDROP_ACCESS_TOKEN=your_token_here
4. Register with Claude
<details> <summary><strong>Claude Desktop (macOS / Windows)</strong></summary>
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"rainkeeper": {
"command": "python",
"args": ["-m", "rainkeeper.server"],
"cwd": "/absolute/path/to/rainkeeper",
"env": {
"RAINDROP_ACCESS_TOKEN": "your_token_here"
}
}
}
}
Restart Claude Desktop after saving. </details>
<details> <summary><strong>Claude CLI</strong></summary>
claude mcp add rainkeeper -- python -m rainkeeper.server
Make sure your .env is in the rainkeeper directory and contains your token.
</details>
5. Verify
Ask Claude: "List my Raindrop collections."
You should see your collections within a few seconds. If you get an auth error, verify the token in your config matches what's in Raindrop Settings → Integrations.
Tools
Rainkeeper exposes 17 tools across five domains. Claude selects the right tool automatically — docstrings are tuned for tool selection accuracy.
Start every session with
list_collections. You need collection IDs to move or file bookmarks, and Raindrop doesn't use predictable names as identifiers.
Collections
| Tool | What it does |
|---|---|
list_collections |
All collections and sub-collections — id, title, parent_id, count |
get_collection |
Single collection detail |
create_collection |
New collection; pass parent_id to nest inside an existing one |
update_collection |
Rename, recolour, or toggle public visibility |
delete_collection |
Delete a collection — contents move to Unsorted |
Bookmarks
| Tool | Key parameters | Notes |
|---|---|---|
list_raindrops |
collection_id, page, per_page, sort |
0 = All, -1 = Unsorted |
get_raindrop |
raindrop_id |
Full detail: tags, note, type, domain |
create_raindrop |
link, collection_id, title, tags, note |
Saves to Unsorted by default |
update_raindrop |
raindrop_id + any field |
Pass collection_id to move |
move_raindrop |
raindrop_id, target_collection_id |
Explicit move — better tool selection than update |
delete_raindrop |
raindrop_id |
Permanent |
Bulk Operations
| Tool | What it does |
|---|---|
bulk_move_raindrops |
Move a list of bookmark IDs to a target collection |
bulk_tag_raindrops |
Apply tags to a list of bookmarks (merges with existing) |
bulk_mark_important |
Star or unstar a list of bookmarks |
bulk_delete_raindrops |
Permanently delete a list of bookmarks |
Search
| Tool | Notes |
|---|---|
search_raindrops |
Searches all bookmarks by default (collection_id=0). Full Raindrop search syntax supported. |
Search syntax:
| Syntax | Example | Finds |
|---|---|---|
| Word | python |
Bookmarks containing "python" |
| Phrase | "machine learning" |
Exact phrase match |
| Tag | #ai |
Bookmarks tagged "ai" |
| Type | type:article |
Articles only (article, video, image, document, audio) |
| Exclude | -draft |
Excludes the word "draft" |
| Domain | site:github.com |
From a specific domain |
| Untagged | # |
Bookmarks with no tags |
| Unsorted | collection:unsorted |
Items in Unsorted |
| Starred | important:true |
Starred / important bookmarks |
Tags
| Tool | What it does |
|---|---|
list_tags |
All tags across your library; pass collection_id to scope |
rename_tag |
Rename a tag everywhere it appears |
delete_tags |
Remove one or more tags from all bookmarks |
Sweep Workflows
Rainkeeper is built for inbox-zero sweeps: work through Unsorted, tag and file everything, leave nothing behind.
"List everything in my Unsorted collection and suggest a collection for each."
"Move all bookmarks tagged #read-later to my Reading collection."
"Search for everything tagged #ai from the last 30 days and list titles and URLs."
"Rename the tag 'ai-tools' to 'ai' across my whole library."
"File the 20 most recent bookmarks into the right collections based on their titles."
"Delete everything in Unsorted older than 6 months."
Architecture
rainkeeper/
├── src/
│ └── rainkeeper/
│ ├── server.py FastMCP app · registers all tools
│ ├── client.py RaindropClient · all HTTP lives here
│ ├── config.py RAINDROP_ACCESS_TOKEN · BASE_URL · headers
│ └── tools/
│ ├── collections.py 5 tools — list · get · create · update · delete
│ ├── raindrops.py 6 tools — list · get · create · update · move · delete
│ ├── bulk.py 4 tools — move · tag · star · delete (batch)
│ ├── search.py 1 tool — full-text + Raindrop syntax
│ └── tags.py 3 tools — list · rename · delete
├── .env.example token template — copy to .env, never commit .env
├── pyproject.toml installable package · rainkeeper CLI entrypoint
└── README.md
Design rule: RaindropClient owns all HTTP. Tools are thin wrappers — no httpx in tool files. The API mapping (including the collection move fix) is enforced in one place and testable independently of the MCP layer.
Known Limitations
The following Raindrop.io API capabilities are not yet exposed. Open a feature request if any matter to your workflow, or start an idea in Discussions.
- Highlights — create and read highlights within a bookmark
- User profile —
GET /user(handy for verifying auth is working) - Imports / exports — bulk HTML bookmark import, export to file
- Sharing — shared collection management
- OAuth 2.0 — multi-user auth; currently personal test token only
Contributing
Rainkeeper is maintained by @simonives. See CONTRIBUTING.md for how to raise bugs, propose features, and fork the project.
License
MIT — see LICENSE.
<div align="center"> <sub>Built with <a href="https://github.com/jlowin/fastmcp">FastMCP</a> · <a href="https://developer.raindrop.io">Raindrop.io REST API v1</a></sub> </div>
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。