# Slant

> An AI-native collaborative writing editor that interviews you, drafts in your voice, and invites other agents to refine your work.

## What Slant Does

Slant helps humans and AI agents collaborate on documents. It has two modes:

**Chat Mode** — Slant interviews you about your topic, researches the web, then generates three structurally different drafts in your voice. You pick the best approach and refine it with AI or human collaborators.

**Manual Mode** — A direct editing environment where you write from scratch, with optional AI agent collaboration via the bridge API.

## What Agents Can Do

Agents connect over the **Slant MCP server** (a workspace-scoped connection — the recommended path) or a **tokenized share link** (single document, no install). Depending on their permission level, they can:

- **Read** — View document content, comments, suggestions, style guides, and query knowledge bases
- **Comment** — Leave comments anchored to specific text
- **Suggest** — Propose text replacements that the author can accept or reject
- **Edit** — Directly modify document content and delete nodes
- **Presence** — Show up in the real-time collaborator bar with status updates (thinking, reading, writing, completed)

## Connect via MCP (recommended)

The richest way to work with Slant is the **Slant MCP server** — one workspace-scoped connection that lets an agent list/search/read/create documents, comment, suggest, directly edit (text, lists, and checklists), search knowledge bases, load voices, publish, and mint share links — every change attributed to the agent.

- **Server URL:** `https://slant.page/api/mcp` (streamable-HTTP MCP)
- **Token:** the human mints one in Slant → **Settings → AI agents** (choosing its permission + scope) and hands it to you.
- **Claude Code:** `claude mcp add --transport http slant https://slant.page/api/mcp --header "Authorization: Bearer <token>"`
- **Any MCP client:** point it at the URL above with an `Authorization: Bearer <token>` header.

A token = **scope** (which docs it reaches: own-docs / a folder / one doc / the whole workspace) × **permission** (view < comment < suggest < edit), plus **author** powers (create / publish / mint links). To reach a document outside your scope — including one in another workspace — the human gives you a share link; pass its token as the `shareToken` argument to any document tool.

## Share-link bridge (one-off, no install)

For quick single-document access without MCP, a human can hand you a tokenized share URL (`/d/<docId>?token=<token>`). Use the bridge API or the installable skill:

- **Full bridge API reference:** https://slant.page/docs.txt
- **Installable skill with bash helpers:** `mkdir -p ~/.claude/skills/slant && curl -fsSL https://slant.page/SKILL.md -o ~/.claude/skills/slant/SKILL.md`
- **Check for skill updates:** the skill's `slant-update` command checks https://slant.page/api/skill-version.
- **Per-client setup (all clients):** https://slant.page/agents.md (human version: https://slant.page/agents)

## Authentication

**MCP (recommended):** an `Authorization: Bearer <token>` header carrying a workspace MCP token (Settings → AI agents).

**Bridge:** a per-document share token, sent as `Authorization: Bearer <token>` (preferred), `x-share-token: <token>`, or `?token=<token>`. Bridge tokens are scoped to a single document with a specific permission level (view, comment, suggest, or edit).