Krill.to MCP Server
Official# @krillto/mcp-server
MCP server for [krill.to](https://krill.to) — search and manage your Twitter/X bookmarks from any AI agent.
Works with **Claude Desktop**, **Claude Code**, **Cursor**, **Windsurf**, **OpenClaw**, and any MCP-compatible client.
## Quick Start
### 1. Get your API key
Sign in to [krill.to](https://krill.to), go to **Settings → API Key**, and generate a key.
### 2. Configure your MCP client
**Claude Desktop** — add to `claude_desktop_config.json`:
```json
{
"mcpServers": {
"krill": {
"command": "npx",
"args": ["-y", "@krillto/mcp-server"],
"env": {
"KRILL_API_KEY": "krill_your_key_here"
}
}
}
}
```
**Cursor** — add to `.cursor/mcp.json`:
```json
{
"mcpServers": {
"krill": {
"command": "npx",
"args": ["-y", "@krillto/mcp-server"],
"env": {
"KRILL_API_KEY": "krill_your_key_here"
}
}
}
}
```
**Claude Code** — add to `.claude/settings.json`:
```json
{
"mcpServers": {
"krill": {
"command": "npx",
"args": ["-y", "@krillto/mcp-server"],
"env": {
"KRILL_API_KEY": "krill_your_key_here"
}
}
}
}
```
That's it. Your AI agent can now search your bookmarks.
## Tools
| Tool | Description |
|------|-------------|
| `search_bookmarks` | Search by keyword, topic, or author (full-text + semantic) |
| `ask_bookmarks` | Ask AI questions about your saved bookmarks |
| `list_collections` | List all your collections |
| `get_collection_bookmarks` | View bookmarks in a collection |
| `create_collection` | Create a new collection |
| `add_to_collection` | Add a bookmark to a collection |
| `update_bookmark` | Update category or tags |
| `add_notes` | Add personal notes to a bookmark |
| `get_highlights` | View all your highlighted passages |
| `delete_bookmark` | Delete a bookmark |
## Examples
Ask your AI agent:
- *"Search my bookmarks about React Server Components"*
- *"What have I saved about AI agents lately?"*
- *"Create a collection called 'Must Read' and add the latest 3 bookmarks"*
- *"Show me my highlights"*
- *"What did @karpathy tweet about that I saved?"*
## What is krill.to?
[Krill.to](https://krill.to) saves your Twitter/X bookmarks, auto-categorizes them with AI, and makes them searchable. Think of it as a second brain for everything you bookmark on X.
- One-click save via Chrome extension
- AI-powered categorization and tagging
- Full-text and semantic search
- Thread unrolling and media capture
- AI chat — ask questions about your bookmarks
## Development
```bash
git clone https://github.com/krillto/mcp-server.git
cd mcp-server
pnpm install
pnpm build
```
Test locally:
```bash
KRILL_API_KEY=krill_your_key node dist/index.js
```
## Environment Variables
| Variable | Required | Description |
|----------|----------|-------------|
| `KRILL_API_KEY` | Yes | Your krill.to API key (starts with `krill_`) |
| `KRILL_BASE_URL` | No | Custom API URL (default: `https://krill.to`) |
## License
MIT
TDQS
Scored across 10 tools
Each tool has a clearly distinct purpose: updating bookmarks, retrieving highlights, deleting bookmarks, searching, asking questions, managing collections, adding notes, and adding to collections. Even search_bookmarks and ask_bookmarks are clearly separated (one returns matches, the other provides an AI-generated answer). No two tools appear to overlap in function.
The verb_noun pattern is consistent with lowercase and underscores, but there are minor deviations: some nouns are singular (update_bookmark, delete_bookmark) while others are plural (search_bookmarks, ask_bookmarks), and add_to_collection uses a prepositional phrase instead of a simple verb+noun. However, the overall style is uniform and predictable.
The tool count of 10 is well within the typical 3-15 range and aligns with the server's purpose of managing Twitter/X bookmarks and collections. Each tool covers a distinct aspect of the domain without excessive redundancy or unnecessary additions.
The tool set covers core bookmark operations (update, delete, search, retrieve via collections) and collection basics (create, list, add). However, there are notable gaps: no way to delete or rename a collection, no way to remove a bookmark from a collection, and no direct method to fetch a single bookmark by ID. These missing operations limit full lifecycle management.