url-manager-mcp
Enables AI agents to manage bookmarks (webpages, articles, videos) on behalf of users, including registration, categorization, tagging, and sharing.
README
URL Manager MCP Server
Agent-first MCP server for managing users' web bookmarks. Agents auto-register, collect, categorize, tag, search, and share URLs — zero manual setup for end users.
📖 For detailed workflows, patterns, and behavioral guidelines, see the URL Manager Skill.
First Conversation (Every Time)
- If no token provided, ask "Do you have a URL Manager account?"
- No → Call
agent_register()to create one (token auto-applied) - Yes → Have the user provide their token, set
FOOTPRINTS_TOKENenv var - Call
my_info()to confirm the connection works
⚠️
agent_register()creates a NEW account every time. Never call it twice — if the user forgot they already have an account, you'll create a duplicate empty one.
Tool Reference
Registration & Identity
| Tool | When to Use |
|---|---|
agent_register() |
User has no account. No params, returns token auto-memorized |
my_info() |
First-conversation connection check, or user asks "Who am I?" |
Bookmarks
| Tool | When to Use | Key Param Source |
|---|---|---|
search_footprints(query) |
User says "find that article about..." | query from user input |
list_footprints(category_id) |
User says "show me my bookmarks" | category_id from list_categories() result |
get_footprint(footprint_id) |
View full details of one bookmark | footprint_id from search or list results |
add_footprint(url, ...) |
User says "save/bookmark this" | url required; category_ids/tags from list_categories() / list_tags() |
update_footprint(id, ...) |
User says "change title/move to another category" | id from search or list results |
Categories & Tags
| Tool | When to Use | Notes |
|---|---|---|
list_categories() |
ALWAYS call before operating on bookmarks — discover existing structure | Returns personal + shared categories. mode=null → personal, mode="cocreate"/"subscribe" → shared |
create_category(name) |
User says "create a new category" | Call list_categories() first to avoid duplicates |
list_tags() |
Before tagging — avoid duplicate tags | Returns existing tags |
Shared Categories
| Tool | When to Use |
|---|---|
create_shared_category(name, mode) |
User says "create a shared collection". mode="cocreate" (editable) / "subscribe" (read-only) |
create_invite_link(shared_category_id) |
User says "send the invite link to..." — id from list_categories() shared entries |
join_shared_category(invite_code) |
User says "I have an invite code" |
add_to_shared_category(sc_id, footprint_id) |
Add an existing bookmark to a shared category |
remove_from_shared_category(sc_id, footprint_id) |
Remove a bookmark from a shared category |
copy_footprint(footprint_id, category_ids) |
Copy from shared category to your personal category |
Batch Operations
| Tool | When to Use |
|---|---|
batch_update_footprints(updates) |
Bulk edit bookmarks — change categories/titles/tags on up to 50 items at once |
Delivery to User
| Tool | When to Use |
|---|---|
agent_magic_link() |
🔑 The delivery loop core. After organizing, generate a link → send to user. They open it to see a card-based interface |
⚠️ Critical Pitfalls
update_footprint: category_ids REPLACES, not appends
# ❌ Wrong: moving bookmark 42 to category 7 loses existing categories 3 and 5
update_footprint(42, category_ids="7")
# ✅ Right: fetch current categories first, then merge
get_footprint(42) → existing categories [3, 5]
update_footprint(42, category_ids="3,5,7")
subscribe mode is READ-ONLY
Writing to a subscribe-mode shared category returns 403. If the user says "I subscribed but can't add anything", explain it's read-only — the creator needs to change it to cocreate.
Rate Limiting
Rapid consecutive calls may trigger HTTP 429. Add short delays between batch operations; on 429, wait a few seconds and retry.
Typical Workflows
New User From Scratch
1. agent_register() → get token (auto-memorized)
2. add_footprint(url="...") × N → save bookmarks one by one
3. list_categories() → understand current structure
4. create_category(name="Learning") → create a category
5. update_footprint(id, category_ids="...") → categorize
6. Tell user: "Done! Open https://ai.ocean94.com to view" (their token-based account)
Returning User — Daily Use
1. my_info() → confirm identity
2. list_categories() + list_tags() → understand current structure
3. search_footprints(query) or list_footprints(category_id) → find targets
4. add_footprint / update_footprint → operate
Create Shared Category
1. create_shared_category(name="Team Knowledge Base", mode="cocreate")
2. create_invite_link(shared_category_id=<returned ID>)
3. Send the invite code to the user → user shares with teammates
4. Teammates' Agents join via join_shared_category(invite_code)
Pairing with Popular MCP Servers
URL Manager excels at saving and organizing. Pair it with tools that excel at discovering and fetching:
Fetch MCP scrapes web → add_footprint() → auto-categorized, permanent, searchable
Firecrawl crawls pages → add_footprint() → organized into cards
Brave Search finds URLs → add_footprint() → one-click save from search results
Agents just pass the upstream MCP's URL + title as params to add_footprint.
Deployment (for Humans)
git clone https://github.com/Piccolo123/url-manager-mcp.git
cd url-manager-mcp && pip install -r requirements.txt
Cherry Studio / Claude Desktop config:
{
"mcpServers": {
"url-manager": {
"command": "python",
"args": ["path/to/server.py"]
}
}
}
If user already has an account, add: "env": {"FOOTPRINTS_TOKEN": "FA_xxx"}
ModelScope: url-manager-mcp
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。