Connect Raindrop.io to your AI assistant with a simple MCP server. Use it to organize, search, and manage bookmarks with natural language.
Raindrop.io now also offers a hosted Streamable HTTP MCP endpoint () in beta for Pro users. Use that when its hosted toolset is sufficient; use this package for local stdio, self-hosted HTTP, and its additional library-management tools.
MCP Server for Raindrop.io bookmark management
2026-11-16
星标
—
工具
—
可用率
—
https://api.raindrop.io/rest/v2/ai/mcp
What it can do
Create, update, and delete collections and bookmarks
Search bookmarks by tags, domain, type, date, and more
Manage tags (list, rename, merge, delete)
Read highlights from bookmarks
Bulk edit bookmarks in a collection
Audit broken links and duplicates, and manage trash
Tools
diagnostics - Server diagnostic information and library health metrics
collection_list - List all collections as a flat list
get_collection_tree - Hierarchical view of collections with full breadcrumb paths
collection_manage - Create, update, or delete collections
bookmark_search - Advanced search with filters, tags, and pagination
bookmark_manage - Create, update, or delete bookmarks
get_raindrop - Fetch a single bookmark by ID
list_raindrops - List bookmarks for a collection with pagination
get_suggestions - AI-powered organization advice (tags/collections) for a URL or bookmark
suggest_tags - Suggest relevant tags from bookmark metadata using AI-assisted analysis
bulk_edit_raindrops - Bulk update, move, or remove bookmarks in a specific collection
tag_manage - Rename, merge, or delete tags
highlight_manage - Create, update, or delete highlights
library_audit - Scan library for broken links, duplicates, and untagged items
empty_trash - Permanently empty the trash (requires confirmation)
Note: For tools adding an MCP server via npx, you must have the RAINDROP_ACCESS_TOKEN environment variable set in your shell or the tool's environment config.
Vercel Skills (npx skills)
This project is compatible with the Vercel Skills system. All necessary manifests are present:
manifest.json: Standard MCP manifest for server definition.
SKILL.md: Standardized skill description for agent discovery.
mcp.json: Root configuration for MCP clients and registries.
To add this server to your local skill directory:
npx skills add adeze/raindrop-mcp --global
To contribute this to the Vercel Skills Registry, submit a Pull Request adding this repository URL to the registry.
Claude Desktop (MCPB)
Download the latest raindrop-mcp.mcpb from the GitHub Release and add it to Claude Desktop:
bun run start:http serves /mcp on 127.0.0.1:3002; set HTTP_PORT to
change the port. HTTP tools use the server's RAINDROP_ACCESS_TOKEN. Host and
Origin checks prevent DNS rebinding but do not authenticate callers. For remote
access, place an authenticated reverse proxy in front of the loopback endpoint.
This repository uses semantic-release as the only supported release flow.
How publishing works
Releases run from pushes to master via .github/workflows/publish.yml after validation passes.
Release runs are serialized; only the release job receives repository write and OIDC permissions.
semantic-release analyzes Conventional Commit messages, computes the next version, updates CHANGELOG.md, tags/releases on GitHub, and publishes npm.
During release preparation, .releaserc.json syncs manifest.json, mcp.json, and gemini-extension.json, then builds raindrop-mcp.mcpb so the GitHub Release includes the bundle.
Pre-release dry-run
In GitHub Actions, select Validate and Publish → Run workflow on master. Leave dry_run enabled to preview a release; disable it to run semantic-release publishing.
This validates semantic version calculation, registry auth, and release pipeline behavior without publishing.
Required secrets
GITHUB_TOKEN is provided by GitHub Actions for release automation.
npm trusted publishing
npm publishing is configured via GitHub Actions OIDC trusted publishing.
Keep the trusted publisher workflow filename set to publish.yml, including manual dry runs.
NPM_TOKEN is not required for the standard CI release path.
Local validation before merge
Use Node 24.15+ and Bun 1.4.3+. The published server requires Node 22.19+.
Vite+ 1.1 manages lint, formatting, type checks, tests, and packaging through
vite.config.ts; Bun remains the package manager.
bun install --frozen-lockfile
bun run check
bun run test
bun run test:coverage
bun run docs
bun run build runs vp pack, preserving build/index.js, build/server.js,
source maps, and bundled dependencies. Tests build first so package checks work
from a clean checkout. bun run dev and bun run dev:http retain Bun's server
watch mode; bunx vp run dev runs the same script. vp dev starts a Vite web
server and is not this MCP server's development command.
The server uses TypeScript 7.0.2. TypeDoc and OpenAPI type generation currently need the TypeScript 6 compiler
API, so tooling/compiler-api keeps its own TypeScript 6.0.3 dependency. Install from the
root; bun run docs still generates the existing documentation.
bun run update:deps refreshes dependencies while keeping coverage aligned with
Vite+'s bundled Vitest 5.0.3. Update those pins together when upgrading Vite+.
The migration removed the generated clearMocks: false setting after the same
61 tests passed with Vitest 5 defaults. Coverage resolves the same 16 files as
Vitest 4; no thresholds or exclusions were relaxed. No tsdown compatibility
settings were needed because the original build used Bun.
The unused Java/Axios generator stack was removed; typed openapi-fetch remains
the shared client. Both current OpenAPI generators tested require the older
TypeScript compiler API: Hey API 0.99.0 fails under TypeScript 7 before CLI help.
The isolated, tested generator avoids changes to authentication, rate limiting
and cache behavior.
Release tooling still pulls braces 3.0.3, affected by
GHSA-vfj7-8cjw-p6xm;
no patched version is published. This dependency is not included in the server
bundle. Do not feed untrusted glob patterns to release tooling.
Breaking changes and reviewed fixes
Deleting bookmarks, collections, tags or highlights, and bulk removal now
requires confirm: true. Cleanup tools keep their existing preview flow.
HTTP binds to 127.0.0.1; OAuth callbacks require matching, unexpired,
single-use browser state. Remote access requires an authenticated proxy.
Bulk moves/updates/removals use the source collection's API endpoint.
Bookmark updates preserve collectionId; omit URL/title when changing only
other fields. Collection 0 means All and cannot be used for bulk mutations;
-1 means Unsorted.
Tags use upstream search operators, preserving pagination and total counts.
Collection paths no longer depend on API ordering; mutations invalidate
affected bookmark, search and collection-count caches.
Permanent upstream errors are not retried. Resource templates use the MCP
SDK; unsupported resource subscriptions are no longer advertised.
MCPB packages contain bundled runtime code rather than development
dependencies. Environment files, logs and local Codex state are excluded.
These fixes address the substance of PR #52
and issues #54/#55.
Older SDK v1/Zod 3 migrations are superseded by this SDK v2 setup. Framework
replacement was unnecessary: existing SDK APIs and shared service helpers
eliminated redundant code. Live Raindrop mutations and OAuth provider login
remain manual validation steps.
Commit message examples
fix: handle empty tag merge payload
feat: add collection path filter
feat!: remove deprecated search parameter
Do not manually bump versions, push release tags, or run manual npm publish commands for normal releases.
📋 Recent Enhancements (v2.4.x)
Smart Organization & Hierarchy
AI Suggestions: New get_suggestions tool provides organizational advice using Raindrop's API and MCP Sampling.
Collection Tree: get_collection_tree tool provides a hierarchical view with full breadcrumb paths.
Bulk Move: Added move operation to bulk_edit_raindrops for efficient library organization.
Pagination Support: Standardized list_raindrops and bookmark_search with pagination for large libraries.
Safety & Quality
Confirmation Logic: Destructive tools (empty_trash, cleanup_collections) now require explicit confirmation.
Standardized Naming: All tools now use consistent snake_case naming conventions.
CI/CD Pipeline: Enhanced GitHub Actions with automated linting, type-checking, and cross-transport tests.
Code Quality: Established ESLint and Prettier configurations for maintainable development.
📋 Previous Enhancements (v2.3.3)
Advanced Cleanup & Library Audit
📋 Previous Enhancements (v2.3.2)
MCP Resource Links Implementation
Modern resource content following current MCP SDK best practices
Efficient data access: tools return lightweight links instead of full payloads
Better performance: clients fetch full bookmark/collection data only when needed
Seamless integration with dynamic resource system (mcp://raindrop/{id})
SDK & API Updates
Updated to the latest supported MCP SDK in this repository
Modern tool registration with improved descriptions
Fixed API endpoints and path parameters
All core tools fully functional
Tool Optimization
Resource-efficient responses for bookmark/collection lists
Dynamic resource access via mcp://collection/{id} and mcp://raindrop/{id}
Better client UX with lighter list payloads
Full MCP compliance with official SDK patterns
Service Layer Improvements
Reduced code through extracted common helpers
Consistent error handling and response processing
Enhanced type safety with generic handlers
Centralized endpoint building
Testing Improvements
Stronger end-to-end coverage for MCP tool execution
Expanded integration tests for real-world client flows
MCP 2.0 Preparation (Bulk Ops)
Laying groundwork for MCP 2.0 bulk-operation workflows and tooling
OAuth (Coming Soon)
OAuth-based auth flow to simplify setup without manual tokens
Note
Apologies to anyone affected by the last couple of builds. Thank you for the patience and reports.
License
This project is licensed under the MIT License - see the LICENSE file for details.