# LinkList

> Grounded sources for agents. LinkList serves source collections you vetted to AI agents over MCP, so
> answers arrive with the source URL and the date it was fetched attached to every passage.

LinkList is grounding infrastructure, not an application you work inside. It has no reading surface of its
own: the agent you already use (Claude Code, Codex, Cursor, or any MCP client) is the interface.

## How it works

1. Add the URLs you trust to a collection. LinkList fetches the full text, chunks it and embeds it.
2. Register the endpoint with your MCP client:
   `claude mcp add --transport http linklist https://linklist.io/mcp`
3. Your agent calls `search_sources` when it needs grounding and answers with citations attached.

## Endpoint

- URL: `POST https://linklist.io/mcp`
- Transport: MCP streamable HTTP, JSON-RPC 2.0, stateless
- Protocol versions: 2025-11-25 (current), 2025-06-18, 2025-03-26 — negotiated at `initialize`
- Auth: `Authorization: Bearer <api key>`
- Tools: `search_sources(query, collection?, max_results?)`, `list_collections()`, `get_source(link_id)`

Every passage returned carries `url`, `title`, `fetched_at`, `link_id` and a relevance score as structured
fields. Citations are data, not prose.

## Retrieval quality

Scores are published and reproducible. The eval harness ships with the product; run
`rails eval:retrieval` against the golden set to reproduce them. Current figures are rendered on
https://linklist.io from the latest real run, including when they fall short of our own 0.85 launch bar.

## Pricing

- Free: 3 collections, 100 sources, 1 key, 100 queries/day
- Pro, $12/month: unlimited collections and sources, 5 keys, 2,000 queries/day, manual re-fetch

## How this differs from adjacent tools

- Memory products (Mem0, Supermemory) store what an agent remembers — implicit, accumulated, unauditable.
  LinkList serves what a person or team explicitly approved, with provenance on every passage.
- NotebookLM is a destination: you work in its interface and it runs Gemini on every tier, with 50 sources
  per notebook on the free plan. LinkList works inside whatever client and model you already chose.

## Pages

- https://linklist.io — overview, live eval figures, setup command
- https://linklist.io/browse — public source collections