Raindrop MCP: Connect Your Bookmarks to Claude

Most guides about getting Raindrop into Claude are out of date, and they are out of date in a specific way. They tell you to pick a community MCP server off GitHub, generate a personal token, and wire it up locally. That was the only route for a long time.

It is not any more. Raindrop ships its own hosted MCP server now, and that changes the answer to this question completely. There is no repo to trust, no local process to babysit, and no export to move. You point your agent at a URL and approve it in a browser.

There is a catch, and there is a bigger catch behind it. The first is that the server is in beta and Pro only. The second is the one nobody selling you a workflow will mention, which is that connecting Raindrop to your agent only helps to the extent that Raindrop is where your saving actually happens.

What Raindrop MCP actually is

MCP, the Model Context Protocol, is the standard that lets an AI assistant call an external tool during a conversation. A Raindrop MCP server is the piece that lets Claude search and manage your Raindrop library without you copying anything into the chat.

Until recently every Raindrop MCP server was a community project wrapping the public API. Several are still around and several still work. The reason to stop caring about them is that Raindrop now documents its own, hosted at https://api.raindrop.io/rest/v2/ai/mcp.

The documentation is specific about the terms. It uses Streamable HTTP, described as the only supported transport. It authenticates with OAuth 2.1, or with the same bearer token the REST API uses if you prefer to skip the browser step. And it is, in Raindrop’s own words, in beta and available to Pro users.

Coverage runs across five areas, and it is wider than the community servers ever managed. Bookmarks (search, fetch content, spot misplaced or mistagged items, create, update, delete). Collections (find, create, update, merge, delete). Tags. Highlights. Plus account-level stats and popular keywords.

Read that list once more with your security hat on. Raindrop notes that read-only tools only fetch data and the rest modify your library. Merging collections and deleting tags are destructive operations, and you are handing them to something that occasionally misunderstands you.

How to connect Raindrop to Claude

Five minutes, assuming you are on Pro.

Confirm the plan first: This is the step that decides whether the rest of the page applies to you. The server is gated to Pro, so a free account will fail at authorization rather than at setup, which is a confusing place to fail.

Add the endpoint to your client: In Claude, Cursor, or whatever you use, add a remote MCP server pointing at https://api.raindrop.io/rest/v2/ai/mcp. Because the transport is Streamable HTTP only, a command and args style config aimed at a local process will not work here. If your client cannot reach a remote HTTP server directly, you need mcp-remote in front of it, the same pattern the Obsidian MCP setups run into.

Approve it in the browser: OAuth opens a tab and asks you to authorize access to your account. Approve it and the tools appear. If you would rather not run the interactive flow, pass your existing API token as an Authorization: Bearer header instead.

Then ask it something hard: Not “list my bookmarks”. Ask for the thing you half remember. That is the only test that tells you anything, and it doubles as an audit of your own library.

You do not need to export anything

This is worth saying plainly, because “raindrop export” and “raindrop mcp” get searched by the same person on the same afternoon and they are not the same task.

The MCP server reads your live account. There is no migration step, no file to produce, no import to run. If you came here to move your collection somewhere, the connection route skips that entirely.

Export is still worth doing, just for different reasons. Raindrop exports to HTML, CSV and TXT, and export works on the free plan. Pro accounts also generate daily backups automatically on days you changed something, keeping up to 30 before the oldest gets overwritten.

So the honest sequence is this. Export because you should have a copy of your own data, not because the agent needs it. Connect because that is what actually puts the library in front of Claude.

The limit nobody puts in the tutorial

Here is the part that decides whether any of this was worth five minutes.

An MCP server can only answer for what is inside the tool it wraps. Connect Raindrop and your agent can search everything you deliberately clipped into Raindrop with a browser extension. That is a real and useful set of things.

It is also, for most people, a minority of what they saved this year.

Think about where saving actually happened. You hit the bookmark icon on a thread in X because it was faster than opening a tab. You saved a Reddit comment that answered a question better than any article. You hit save on a LinkedIn post you meant to come back to. None of those touched Raindrop. They are sitting inside three platforms with no export worth the name and, in two cases, a hard cap on how far back the list even goes.

The failure mode is not that the agent says “I do not know”. It is that the agent confidently answers from half your library and you have no way to tell that the better source was the one it could not see.

So is Raindrop still the right tool?

Genuinely, often yes. This is not the part where I tell you to switch.

Raindrop is a good browser-first bookmark manager with an unusually generous free tier, and the Pro plan at roughly $28 a year is one of the better value subscriptions in this category. Now that it has a first-party MCP server, it also clears the bar that mattered most for anyone working inside an agent. If your saving genuinely happens in the browser, connect it and stop reading.

The case for looking elsewhere is narrower than the alternatives lists suggest, and it is not about features. It is about where you save. If you looked at that table and most of your year is in the bottom three rows, then the tool is not the problem and switching bookmark managers will not fix it. You have a capture gap, not a storage gap.

We keep a ranked breakdown at best Raindrop alternatives if you want the shopping version of this question. This page is the procedure. That one is the comparison, and they answer different questions on purpose.

Where ContextBolt fits, and where it does not

Straight answer, because this is our site and you should be able to discount it accordingly.

ContextBolt does not import your Raindrop library. There is no CSV route and there is not going to be one. A Raindrop export is a list of links and titles, and the thing ContextBolt stores is the full post, so the shape is wrong. Anyone offering you a one-click migration between the two is describing something that loses most of what makes either useful.

What it does instead is the bottom three rows of that table. It is a Chrome extension that saves any web page with its text and captures what you bookmark on X, Reddit and LinkedIn, tags each save with a topic automatically, and groups them so you can filter. The free Basic tier holds 150 saves. Pro is $6 a month for unlimited saves, cloud sync, and your own MCP endpoint, which is where search by meaning rather than by keyword comes in.

So the honest recommendation is the one Raindrop’s own community reached years ago. Run both. Raindrop for what you clip in a browser, connected over its official MCP. Something that speaks the social platforms for the saves that never reach a browser at all. Two endpoints, one agent, and no gap where you cannot tell what it missed.

If you want the version of this argument that is not about Raindrop specifically, why social bookmarks disappear is the longer case.