Claude Code workflow management

The Claude Code + project management gap

Claude Code is remarkably capable at the implementation layer. Given a clear brief, it can explore a codebase, write tests, implement a feature, and open a pull request — often with minimal back-and-forth. Teams using Claude Code regularly report 3–5× improvements in implementation throughput for well-specified work.

But "well-specified" is doing a lot of work in that sentence. Claude Code doesn't know what your team is working on. It doesn't know which task is highest priority, what the acceptance criteria are for the current sprint (or flow cycle), or what's already been decided in previous planning conversations. Every session starts with context that has to be re-established manually.

And when the implementation is done, Claude Code can't close the loop on the project board. The card stays in "in progress" until a human updates it. If Claude discovers a bug while working on something else, there's no way for it to create a tracked work item without leaving the terminal. The AI does the work, but the project management side stays manual.

How Glacier's MCP server bridges the gap

Glacier exposes an MCP server that Claude Code can connect to directly. Once connected, Claude has access to a set of typed tools for reading and writing your project board:

These tools give Claude Code a complete read-write interface to your project board. It can look up the card it's working on, read the linked brief, implement the feature, update the card status, and create a doc summarising what was done — all without leaving the terminal, and without any manual board updates.

A practical workflow: brief → cards → implementation → status update

1. 1

   Start a Claude Code session with context

   Claude reads your board via MCP, lists the cards in the active column, and asks which one to work on — or you point it to a specific card ID. It fetches the card and linked doc to get the full brief.

2. 2

   Implement with full context

   Claude has the acceptance criteria, technical notes, and any parent card context from the doc. It explores the codebase, writes tests, and implements the feature. If it discovers a related bug, it creates a new card on the board so it's tracked — rather than silently fixing it or losing it.

3. 3

   Update status without leaving the terminal

   When the PR is ready, Claude updates the card status — moving it to the review column via MCP. The board reflects the actual state of the work without any manual updates. A reviewer can see the PR link in the card and the card in the "in review" column.

4. 4

   Close the loop on merge

   After review and merge, the card moves to done. Cycle time is recorded. The board accurately reflects what shipped, when, and how long it took — with no manual data entry.

Setting up Glacier MCP with Claude Code

Add the following to your Claude Code MCP configuration file (~/.claude/mcp.json or your project's .mcp.json):

{
  "mcpServers": {
    "glacier": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.getglacier.ai/mcp"],
      "env": {
        "GLACIER_API_KEY": "your-api-key-here"
      }
    }
  }
}

Your API key is available in your Glacier workspace settings under Settings → MCP. Once connected, Claude Code can use all read and write tools immediately — no additional configuration needed.

For the full setup walkthrough, including a sample CLAUDE.md configuration and example tool invocations, see the MCP connection guide andMCP workflows documentation.