DEV Community

Kyurations
Kyurations

Posted on

Give Claude and Cursor your bookmarks: the Kyurations MCP server

#mcp #ai #showdev #productivity

Most of us save a lot of links. For years, our AI tools could not see any of them. When you ask Claude about something you read, it answers from the open web, not from the sources you picked. So we added a Model Context Protocol (MCP) server to Kyurations, our bookmarking app.

This post explains what the server does, how to connect it, and what you can ask it.

What the server does

Kyurations speaks MCP. That means Claude, Cursor and any agent that talks the protocol can search, read and help organise the library you have curated.

The endpoint is:

https://www.kyurations.in/api/mcp
Enter fullscreen mode Exit fullscreen mode

The tools are in two groups.

Read tools. These never leave your library.

  • Search the library. Semantic and keyword search across everything you saved. Every result comes back with its link.
  • Filter by facet. Exact lookups by platform, type, author, tag, date or a named list.
  • Count and break down. For example: how many videos you saved, what you save most about, and how much sits in each list.
  • Read in full. Pull the full text or transcript of one item, with its source URL, author and tags.
  • See your lists. Every list with its id, its size and its public link, if you shared it.

Write tools. These are scoped to your token and reversible, and your client approves each call.

  • Create and fill lists. Make a list, seed it with items, then add or remove things.
  • Save new links. Give it a URL. It extracts the title, tags and a summary, and checks for duplicates against what you already keep.
  • Write notes. Keep a thought with no link, straight into the library or a list.
  • Edit what is saved. Fix a title, rewrite a summary or set a priority. Only the field you name changes.
  • Tag and retag. Add or remove tags across items.
  • Publish a list. Make a list public and get the share link back, or make it private again.
  • Delete, recoverably. Delete is a soft delete to the trash, and you can restore from it. The tool is marked destructive, so your client asks you first.

We kept the set small on purpose. Each tool does one clear job, which makes it easier for the model to choose the right one.

How to connect it

Step 1: Generate a token.

  • In Kyurations, open Settings.
  • Open "Connect Kyuri to your AI tools" and generate a token.
  • The token starts with kyur_. You see it only once, so copy it.
  • You can revoke it at any time under Connected apps.

Step 2: Add the server to your client.

The easy way: the /mcp page has a set-up prompt. Copy it and paste it into Claude, ChatGPT, Cursor or whatever you use. The prompt covers every client. Your agent works out which config to edit, and it checks the connection when it is done. Replace kyur_YOUR_TOKEN with your token, or copy the prompt from Settings with the token already in place.

The manual way, for Claude Desktop: open Settings, then Developer, then Edit Config. Add this block and put in your own token:

{
  "mcpServers": {
    "kyurations": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://www.kyurations.in/api/mcp",
        "--header", "Authorization: Bearer kyur_YOUR_TOKEN"
      ]
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

For Cursor, use the set-up prompt from the /mcp page. It edits the right config for you.

Want to test before you wire it up? Point the MCP Inspector at the same endpoint with an Authorization: Bearer header:

npx @modelcontextprotocol/inspector
Enter fullscreen mode Exit fullscreen mode

Step 3: Restart and ask.

Restart your client. Kyurations shows up as a tool. The answers come from your library, and they cite the items they used.

What you can ask

Here are a few prompts that match the tools above:

  • "What did I save about rate limiting? Give me the links."
  • "How many videos are in my Interaction Patterns list?"
  • "Read the full transcript of that talk on fluid UI and give me the three main points."
  • "Make a list called Onboarding References and add everything I saved about onboarding flows."
  • "Save this URL and tag it with design."
  • "Make my Reading List public and give me the link."

The read questions are where this feels different. The model stops guessing from the open web. It answers from sources you already chose to trust.

Safety, in short

  • Your token sets the scope.
  • Every change is reversible.
  • Your client approves each call.
  • Delete goes to the trash, not into the void.

If you try it, tell us which tool you use most and which one is missing. Docs and token setup: https://www.kyurations.in/mcp

Top comments (0)