Your workspace.
Claude’s workspace.

Kanban board. Linked docs. MCP-first.
Built for small teams where AI is already in the daily workflow.

Flow-basedMCP-nativeClaude as collaborator

Connect Claude →Already have access? Sign in →

This is our actual board.

app.getglacier.ai

Backlog

File attachments (R2 storage)

platform

@mentions with email notifications

feature

In Progress

2/2

Landing page redesign

Role-based MCP permissions

mcp

In Review

1/2

Resend invitation

platform

Done

Threaded card comments

feature

Command palette (⌘K)

The problem

Every workspace was built before AI joined the team.

Linear is fast but siloed. Notion is flexible but slow. Neither assumes Claude is reading your board.

What you see: a clean Kanban board and a linked doc editor.

waterline

What’s underneath: an MCP server that exposes your entire workspace to Claude.

What’s built

Not a prototype. A working product.

300+ issues shipped. Three teams running on it daily. Here’s what’s already built.

Kanban board

Column types (backlog / active / waiting / done)
WIP limits with visual indicators
Cycle time and throughput metrics
Flight levels (opt-in per project)
Drag-and-drop with keyboard shortcuts
Full-text search, command palette (⌘K)

Cards

Card hierarchy (parent / sub-items)
Subtask checklists
Threaded comments with markdown
GitHub issue and PR linking
Priority levels and assignees
Custom label taxonomy

Docs and knowledge

ProseMirror rich text editor
Doc hierarchy (parent / child)
Card-linked documentation
Project-scoped and workspace-scoped
Table support
Markdown round-trip conversion

Agent suggestions & inline review

Claude proposes doc changes via MCP
Inline diff: strikethrough + replacement in flow
Rationale visible in hover popover and side panel
Accept or reject per suggestion, or bulk review
Agent vs human attribution on every annotation
Agent Edit Policy (workspace / project / doc level)

Platform

Multi-workspace, multi-project
Member invitation with email + token
Owner / Admin / Member / Viewer roles
Dark and light theme
OAuth 2.1 (Claude.ai/Code) + API key (Desktop)
Role-based MCP tool permissions

MCP-first

Claude doesn’t summarize your board. It uses it.

One authorization gives Claude full workspace access — read and write. Not a plugin, not a wrapper. Native MCP protocol, built into the product from day one. Claude Code, Claude Desktop, Claude.ai, or any MCP-compatible agent connects and starts working.

Read

Write

list_workspaces

list_projects

list_columns

list_cards

get_card

get_card_children

get_doc

list_docs

list_labels

list_members

get_flow_metrics

list_card_comments

list_annotations

create_card

create_cards

update_card

create_doc

update_doc

link_doc_to_card

set_card_labels

update_subtasks

link_card_to_github

add_card_comment

create_column

update_column

propose_suggestion

create_comment

reply_to_annotation

resolve_annotation

reject_annotation

OAuth 2.1 + PKCE

RFC 8414-compliant. Claude.ai and Claude Code sign in with one browser click — no API key, tokens refresh automatically.

API key

For Claude Desktop, Cursor, or CI. Generate from Account Settings, paste once.

Role-gated

Write tools require board:write permission. Viewers get read-only.

Connect Claude in 2 minutes →

Comparison

How Glacier compares.

Linear, Notion, and Atlassian all have MCP servers now — bolted onto architectures that predate the protocol by years. Glacier was designed around MCP from day one: OAuth 2.1 + PKCE auth, role-gated tools, flow metrics, and a unified board + docs surface built for the way teams actually work with Claude.

Capability	Glacier	Linear	Notion	Outline
Kanban / flow board	Native	Issues	DB view	—
Linked docs	Card-linked	Minimal	Core	Core
MCP server	Native (day 1)	Retrofitted	Retrofitted	Community
MCP read + write	Full	Full	Full	Community
OAuth 2.1 + PKCE (MCP auth)	RFC-native	API key	API key	—
Role-based MCP permissions	Native	—	—	—
Flow metrics (cycle time + throughput)	Built-in	—	—	—
Board + docs in one MCP surface	Unified	Issues only	Pages only	Docs only
WIP limits	Built-in	—	—	—
Card comments	Threaded	Threaded	Inline	—
AI doc suggestions (inline review)	Native	—	AI blocks (no diff)	—
AI as collaborator	First-class	MCP + agent	MCP + AI blocks	—
Setup time	Minutes	Fast	High	Simple

Accurate as of April 2026. Linear MCP: launched May 2025, expanded Feb 2026. Notion MCP: hosted server, full read + write.

Who it’s for

Built for one kind of team.

Small technical collectives — 2 to 10 people — where AI is already in the daily workflow. Teams where Claude Code is open next to the terminal. Where context moves faster than standups. Where the bottleneck isn’t headcount — it’s information architecture.

Built for ourselves. Used across three teams and five active products at Black Bear Studio.

Private beta

We build on what we ship.

Black Bear Studio runs on Glacier internally — across three teams and five active products. Every friction point becomes a fix. Every missing feature becomes a card on our own board.

Private beta — free for early teams willing to use it and share honest feedback.

Your workspace.Claude’s workspace.