How do I connect Glacier to Claude Desktop?

Generate an API key in Glacier under Account Settings → MCP API Key, then add Glacier to your claude_desktop_config.json under mcpServers using the MCP server URL https://getglacier.ai/api/mcp and your key as the auth token.

How do I connect Glacier to Claude.ai?

Go to Claude.ai Settings → Connectors, click Add connector, and paste https://getglacier.ai/api/mcp. Claude.ai will redirect you to Glacier to authorise via OAuth 2.1. No API key needed.

What can Claude do with Glacier via MCP?

Claude can read and write your entire workspace: list projects and columns, create and update cards, read and write docs, add comments, fetch flow metrics, and manage labels and members — all using natural language.

Is the Glacier MCP server free to use?

The MCP server is included with every Glacier workspace. API keys and OAuth access are available to all workspace members.

Why does Claude Desktop show no Glacier tools after setup?

Make sure your claude_desktop_config.json is valid JSON (no trailing commas), the API key is correct, and you have fully quit and restarted Claude Desktop — not just closed the window.

What is MCP?

Model Context Protocol (MCP) is an open standard that lets AI assistants like Claude read from and write to external tools — databases, project trackers, documentation systems — without copy-pasting or manual context switching.

Think of it as a universal connector between AI and your workspace. Instead of switching between Claude and your project board, the AI comes to you: aware of what's in progress, able to create cards, update statuses, and draft docs directly.

How MCP changes project management

Before MCP, using an AI assistant with a project management tool looked like this:

Open your project board
Copy a card title and description
Switch to Claude
Paste the context
Get a response
Copy the response
Switch back to your board and paste it in

That's seven steps for every interaction. MCP collapses it to one prompt.

With an MCP-connected workspace, you can ask:

"What's currently in progress? Flag anything that looks blocked."

And Claude will check your board, read card descriptions, and give you a real answer — without you manually providing any context.

What MCP actually does

MCP defines a client-server architecture:

MCP server — a process that exposes tools and resources from an application (your project tracker, your docs, your database)
MCP client — the AI host (Claude.ai, Claude Desktop, Claude Code) that connects to those servers and calls tools on your behalf

The server declares what it can do — list cards, create a card, read a document, update a status — and the AI decides when and how to call those tools based on your prompts.

Tools can be read-only (list cards, get metrics) or write (create card, move column, update description). Glacier exposes both.

Why Glacier is built MCP-first

Most tools bolt MCP on as an integration. Glacier was designed with it as a core interface.

That means:

Card descriptions as context. When you ask Claude about a piece of work, it reads the actual card — description, labels, assignee, current column. Not a stale summary you pasted.

Doc conversion. Claude can read a doc linked to a card, summarise it, or draft a new spec from a card description. The content lives in your workspace, not in a chat window.

Workspace scoping. Every MCP call is scoped to your workspace. Claude can only see and touch the projects and cards you have access to. Multi-tenant by design.

Guarded writes. Creating cards, moving cards, and updating descriptions require explicit tool calls. Claude doesn't modify your board unless you ask it to.

Flow metrics. Claude can query cycle time, throughput, and WIP data — not just read cards, but reason about the health of your workflow.

The protocol is open

MCP is not proprietary to Anthropic or to Glacier. It's an open standard that any AI client can implement. Claude supports it natively, and support is growing across other AI tools.