@glie/mcp-polaris

A Model Context Protocol (stdio) server that exposes Polaris — glie's ops platform — as native tools inside Claude Code, Cursor, or any MCP client. You authenticate as yourself with a Personal Access Token, and the agent can list/transition tasks, validate approval stacks, and approve work without leaving the editor.

Quick start

Add this to your project's .mcp.json (or your Claude Code config). The token is read from the environment, so nothing secret is committed:

{
  "mcpServers": {
    "polaris": {
      "command": "npx",
      "args": ["-y", "@glie/mcp-polaris@latest"],
      "env": {
        "POLARIS_URL": "https://polaris.glie.ai",
        "POLARIS_TOKEN": "${POLARIS_TOKEN}"
      }
    }
  }
}

Then export your token and start your editor from that shell:

export POLARIS_TOKEN='polaris_h_…'   # your Personal Access Token

Getting a token

Open https://polaris.glie.ai, click the gear (top-right) → /user/settings → Personal access tokens → create one. The secret is shown once — copy it. The same token authenticates this server and the polaris CLI as you.

Configuration

| Env var | Default | Notes | | --------------- | --------------------------- | -------------------------------------- | | POLARIS_URL | http://127.0.0.1:7843 | Set to https://polaris.glie.ai. | | POLARIS_TOKEN | (none) | PAT. Read tools work anonymously; writes need it. |

Tools

polaris.me, polaris.project.*, polaris.milestone.* (incl. update to refine a spec), polaris.task.* (incl. update, transition, approve), polaris.stack.* (list / get / release for the approval Release Candidate flow), polaris.agent.*, polaris.claude_md.get, polaris.framework.list, plus read-only context from the feed: polaris.activity.list (what happened) and polaris.chat.get / polaris.chat.messages (comments agents left on a scope).

Run tools/list against the server for the full, current set.

License

Apache-2.0