URL Manager MCP Server

English | 简体中文

Deliver results as beautiful cards, not raw link dumps. A Model Context Protocol server for saving, organizing, searching, and sharing web resources — cross-device sync, categories, tags, full-text search, batch operations, and team sharing. 21 tools with auto-registration so no manual setup is required.

📖 Usage patterns and best practices → URL Manager Skill

What This Tool Gives Humans

The content human users want to save is everywhere — a YouTube workout video, an Amazon gear link, a Substack training plan — scattered across platforms with no connection.

URL Manager fixes this. Paste any link from any platform. AI auto-identifies the content and suggests a category — confirm and it's a footprint. All saves flow into one platform-agnostic library, organized and always findable. Then share in one click — hand your curated knowledge base to your team, and everyone stays in sync.

System Concepts

Footprint (the fundamental unit)

A structured, searchable record — a web link, a plain-text note, an idea, or anything worth saving.

A footprint can belong to multiple categories simultaneously.

Category (a named label)

Like a folder, but a footprint can be in several at once.

Category Set (a workspace)

A container that groups related categories. Every user starts with "My Categories" (personal) and "Shared Categories" (shared container).

Data hierarchy

Category Sets (workspaces)
  └── Categories (labels like "Shopping", "Learning")
        └── Footprints
             └── Tags (free-form keywords)

Personal vs Shared

Cocreate — everyone adds/removes footprints. Subscribe — read-only for members (writing returns 403).

Tools

Registration & Identity

• agent_register() Create a new account. No parameters. Token is auto-applied for all subsequent calls. ⚠️ Call once only — each invocation creates a fresh account.

• my_info() Verify connection and token validity. Returns username and membership status.

Bookmarks

• search_footprints(query, limit, offset) Full-text search across titles, descriptions, and URLs.

  • query (required) — Search keywords
  • limit — Results per page (default 10, max 100)
  • offset — Pagination offset (default 0)

• list_footprints(category_id, limit, offset) List bookmarks by category. category_id=0 returns all.

  • limit — Results per page (default 20, max 100)
  • offset — Pagination offset (default 0)

• get_footprint(footprint_id) Get full details of a single bookmark.

  • footprint_id (required) — From list_footprints or search_footprints results (field id)

• add_footprint(url, title, description, category_ids, tag_names) Add a new bookmark. Call list_categories() and list_tags() first to discover existing structure.

  • url (required) — Web page URL
  • title — Leave empty to auto-extract from the page
  • description — Summary or notes
  • category_ids — Comma-separated IDs, e.g. "1,3"
  • tag_names — Comma-separated names, e.g. "AI,tutorial"

• update_footprint(footprint_id, title, description, category_ids, tag_names) Update a bookmark. Omitted fields stay unchanged. ⚠️ category_ids replaces the entire list — not append. Call get_footprint() first, then merge IDs.

  • footprint_id (required) — From search or list results

Categories & Tags

• list_categories() List all categories (personal + shared). Returns id, name, and mode fields. mode=null → personal; mode="cocreate"/"subscribe" → shared.

• create_category(name, category_set_id) Create a new category. Check list_categories() first to avoid duplicates.

  • name (required) — Category name
  • category_set_id — Parent category set (0 = default)

• list_tags() List all tags used by this account.

• list_content_types() List all content types the user has used (e.g. article, video, image), ordered most-used first. Use before adding to pick a consistent content_type.

Category Sets

• list_category_sets() List all category sets.

• create_category_set(name) Create a new category set (a container of categories).

  • name (required) — Category set name

Shared Categories

• create_shared_category(name, mode, description) Create a shared category for team collaboration.

  • name (required)
  • mode (required) — "cocreate" (multiple editors) or "subscribe" (read-only)
  • description — Optional description ⚠️ In subscribe mode, adding bookmarks returns 403. Use "cocreate" for editable collaboration.

• create_invite_link(shared_category_id, duration_hours) Generate an invite link for others to join.

  • shared_category_id (required) — From list_categories() (shared entries)
  • duration_hours — Default 24

• join_shared_category(invite_code) Join a shared category by invite code.

  • invite_code (required) — 8-character code from the invite link

• add_to_shared_category(shared_category_id, footprint_id) Add one of your own bookmarks to a shared category.

  • Both parameters required

• remove_from_shared_category(shared_category_id, footprint_id) Remove a bookmark from a shared category. Does not delete the bookmark itself.

  • Both parameters required

• copy_footprint(footprint_id, category_ids) Copy a bookmark from a shared category into your personal collection.

  • Both parameters required

Batch & Delivery

• batch_update_footprints(updates) Bulk edit up to 50 bookmarks at once.

  • updates (required) — JSON string: [{"id":"...", "title":"New Title", "category_ids":"1,3"}, ...] Each object may contain title, description, category_ids, tag_names; id is required.

• agent_magic_link() 🔑 The delivery loop core. After organizing, generate a link → send to user. They click to see a card-based interface with all their organized bookmarks. Valid for 30 days, reusable.

Workflows

New User — Zero Setup

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. agent_magic_link() → "Done! View your collection here → [link]"

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
5. agent_magic_link() → deliver results

Create Shared Category

1. create_shared_category(name="Team KB", mode="cocreate")
2. create_invite_link(shared_category_id=<returned ID>)
3. Send invite code to user → user shares with teammates
4. Teammates' agents join via join_shared_category(invite_code)

⚠️ Critical Pitfalls

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.

NEVER call agent_register twice

Each call creates a fresh empty account. Always check for a saved token first. If unsure, call my_info() to verify the current token is valid.

Rate Limiting

Rapid consecutive calls may trigger HTTP 429. Add short delays between batch operations; on 429, wait a few seconds and retry.

No member management via API

Inviting or removing members from shared categories requires the web UI at https://ai.ocean94.com. You cannot do this programmatically.

Guidelines

Always

• Auto-register silently — never interrupt the user for account setup
• Search before listing — use search_footprints for targeted queries instead of dumping everything
• Discover before creating — call list_categories() and list_tags() before adding to avoid duplicates
• Deliver with magic link — after organizing, always generate and share a link

Confirm before

• Removing bookmark-category associations (irreversible)
• Clearing tags
• Modifying cocreate shared categories (affects others)
• Removing bookmarks from shared categories (other members lose access)

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.

Installation

git clone https://github.com/Piccolo123/url-manager-mcp.git
cd url-manager-mcp
pip install -r requirements.txt

Or from PyPI:

uvx url-manager-mcp   # or: pip install url-manager-mcp

Prerequisites

• Python 3.10+
• Network access to https://ai.ocean94.com

DeepSeek Harness (dsh)

Add to your profile's cordis.patch.yml (e.g. ~/.dsh/profiles/headless/cordis.patch.yml or the web profile):

- insert:
    - id: mcp-url-manager
      name: '@deepseek-ai/dsh-mcp-client'
      config:
        serverName: url_manager
        transport: stdio
        command: uvx
        args: ['url-manager-mcp']
        env:
          FOOTPRINTS_ENDPOINT: 'https://ai.ocean94.com'

All 21 tools appear as mcp__url_manager__* (add_footprint, search_footprints, list_categories, agent_magic_link, …). The model auto-registers on first use — no API key needed. The companion skill (Piccolo123/url-manager) is also auto-discovered by dsh when cloned into ~/.dsh/skills/.

Configuration

Claude Desktop / Claude Code

{
  "mcpServers": {
    "url-manager": {
      "command": "python",
      "args": ["path/to/url-manager-mcp/server.py"]
    }
  }
}

If the user has an existing account:

{
  "mcpServers": {
    "url-manager": {
      "command": "python",
      "args": ["path/to/url-manager-mcp/server.py"],
      "env": {
        "FOOTPRINTS_TOKEN": "FA_xxxxxxxxxxxx"
      }
    }
  }
}

Cursor / Windsurf / Cherry Studio

Same JSON structure as above. Works with any MCP-compatible client supporting STDIO transport.

Other Clients

This server supports both STDIO (default) and Streamable HTTP transports:

# STDIO (default)
python server.py

# Streamable HTTP (for Docker / Glama / hosted environments)
python server.py --http

Multi-user remote mode (WorkBuddy / hosted connectors)

Over Streamable HTTP (--http) for a multi-user platform such as the WorkBuddy connector, the server ignores FOOTPRINTS_TOKEN. Every request must carry the caller's own token in the Authorization header (Bearer FA_xxx); it is passed through to the URL Manager API for that user only, so concurrent users stay isolated. Requests without a token get a friendly missing_access_token error pointing the user to https://ai.ocean94.com → 个人中心 → 接入Agent → 访问令牌, and agent_register() is disabled in this mode (it would only create orphan accounts). Single-user STDIO deployments (python server.py, no --http) keep the original FOOTPRINTS_TOKEN / auto-register behaviour unchanged.

Deployment

Docker

docker build -t url-manager-mcp .
docker run -e FOOTPRINTS_TOKEN="FA_xxx" url-manager-mcp

ModelScope

One-click hosted deployment: url-manager-mcp

Why URL Manager

Browser bookmarks are flat lists with no organization, no search, and no sharing. URL Manager adds:

• Categories, category sets, and tags — Hierarchical organization
• Full-text search — Find anything across all titles, descriptions, and URLs
• Cross-device sync — Save on one device, access on all
• Batch management — Sort and organize hundreds of links at once
• Team sharing — Co-editing and read-only shared collections with invite links
• Card-based delivery — Send organized collections as a polished interface, not raw URLs