Rdrop for Raindrop.io

What rdrop for Raindrop.io does

rdrop for Raindrop.io gives an MCP-compatible agent a local interface to a Raindrop.io bookmark library. It is intended for finding saved links, examining collection trees and existing tags, identifying cleanup candidates, and preparing organization plans. The default MCP mode is read-only, so an agent can inspect data and plan changes without immediately modifying the account.

The project also includes the rdrop command-line tool. Developers can use it to search bookmarks, manage collections, edit entries, export library data, work with highlights, and run cleanup or organization workflows without an AI client.

How it works

The MCP server runs locally over stdio and uses a Raindrop.io token stored in the user’s configuration. Its shared command implementation supports bookmark and library reads, while planning commands can create a proposed set of tag and collection changes for review.

A typical organization workflow starts with a cleanup report and a selected bookmark context. The resulting context includes existing tag names, collection paths, and bookmark metadata. A proposal is then validated against that context before a reviewed plan is produced. Plans can preserve existing tags, combine tag changes with collection moves, and identify items that need review.

The CLI can inspect MCP tool schemas locally with commands such as rdrop mcp tools --search and rdrop mcp tools --name. rdrop mcp serve --official can additionally include Raindrop’s hosted AI reads, subject to that extension’s limits and availability.

Setup and configuration

Install with Homebrew, npm, a release binary, or npx. The documented npm path requires Node 22 or newer. After installation, create a token in Raindrop.io’s integration settings and place it in ~/.config/rdrop/env:

RAINDROP_TOKEN=your_token_here

The documentation recommends permissions of 700 for the configuration directory and 600 for the environment file. Run rdrop doctor to check installation and authentication. Keep the token out of chat and source control.

For an MCP client, start the local server with rdrop mcp serve. The repository also documents connection flows for Codex, Claude, and other local MCP clients, including a plugin that bundles a read-only MCP connection and an organizing skill.

Tools and capabilities

The rdrop for Raindrop.io MCP server supports workflows built around:

• Searching saved bookmarks and retrieving bookmark context.
• Inspecting collections, collection paths, and existing tags.
• Finding cleanup candidates and reviewing organization suggestions.
• Discovering available MCP tools and inspecting their schemas offline.
• Producing validated tag and collection plans before any write step.
• Including Raindrop’s official hosted AI read tools when explicitly enabled.

The command-line interface extends these workflows with JSON output, collection trees, exports, highlight management, edits, dry runs, and approved plan application. The project can also run a direct model loop with hosted Responses APIs or local Chat Completions servers, but those model connections are separate from the default read-only MCP operation.

Limitations and notes

Writes are not enabled by default through MCP. The documented organization workflow requires an explicit reviewed plan and recommends a dry run before applying it. Unknown tags or collection destinations fail validation, and new tags require explicit naming rules. Account state, digests, and snapshots are checked before writing; concurrent edits or partial failures require inspection and a fresh plan.

The integration focuses on Raindrop.io storage. Model providers and proposal formats are provider-neutral, while Obsidian and other note sources are identified as future options. Verification described by the project is a smoke check rather than an accuracy benchmark, and some live inference paths require provider credentials.