# ember MCP server

> Provides coding agents with calibrated probability scores to help them decide when to ask questions or take actions.

- Repository: https://github.com/shapeandshare/ember
- Page: https://hostedmcp.org/servers/shapeandshare/ember
- Transport: stdio
- Stars: 0
- License: MIT
- Latest version: 0.16.0

## Install

### Claude Code

```sh
claude mcp add --env EMBER_AUTH_TOKEN=<YOUR_EMBER_AUTH_TOKEN> ember -- uvx ember-advise
```

### Cursor (`~/.cursor/mcp.json`)

```json
{
  "mcpServers": {
    "ember": {
      "command": "uvx",
      "args": [
        "ember-advise"
      ],
      "env": {
        "EMBER_AUTH_TOKEN": "<YOUR_EMBER_AUTH_TOKEN>"
      }
    }
  }
}
```

### VS Code (`.vscode/mcp.json`)

```json
{
  "servers": {
    "ember": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "ember-advise"
      ],
      "env": {
        "EMBER_AUTH_TOKEN": "<YOUR_EMBER_AUTH_TOKEN>"
      }
    }
  }
}
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "ember": {
      "command": "uvx",
      "args": [
        "ember-advise"
      ],
      "env": {
        "EMBER_AUTH_TOKEN": "<YOUR_EMBER_AUTH_TOKEN>"
      }
    }
  }
}
```

### Codex (`~/.codex/config.toml`)

```toml
[mcp_servers.ember]
command = "uvx"
args = ["ember-advise"]

[mcp_servers.ember.env]
EMBER_AUTH_TOKEN = "<YOUR_EMBER_AUTH_TOKEN>"
```

### Gemini CLI (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "ember": {
      "command": "uvx",
      "args": [
        "ember-advise"
      ],
      "env": {
        "EMBER_AUTH_TOKEN": "<YOUR_EMBER_AUTH_TOKEN>"
      }
    }
  }
}
```

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `EMBER_AUTH_TOKEN` | no | yes | Credential for a remote inference endpoint |
| `EMBER_AUTOSTART` | no | no |  |
| `EMBER_DEVICE` | no | no |  |
| `EMBER_HOST` | no | no |  |
| `EMBER_MODEL_S3_URI` | no | no |  |
| `EMBER_SERVER_URL` | no | no | Inference endpoint; defaults to the local model server at http://127.0.0.1:8765 |

## FAQ

### What does ember do for coding agents?

It provides a quick second opinion in the form of a calibrated probability (e.g., '93% likely this needs review') to help agents decide whether to act or ask a question.

### Where can the model run?

You can run it locally on an Apple Silicon Mac, on a self-hosted server (Mac or NVIDIA GPU), or on a hosted server provided by a third party.

### What are the hardware requirements for running it locally on a Mac?

You need an Apple Silicon Mac with at least 32 GB of unified memory for the 'flash' model or 96 GB for the 'full' model.