# ember

> ember is an open-source MCP server for ai models and llm tooling. Calibrated yes/no and multiple-choice odds that help coding agents decide ask or act. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** macOS, self-hosted, CLI
- **Website:** https://shapeandshare.github.io/ember/
- **Repository:** https://github.com/shapeandshare/ember
- **MCP Registry name:** io.github.shapeandshare/ember
- **Categories:** AI models and LLM tooling, developer tools
- **Listing updated:** 2026-11-15

## About

ember answers a narrow question: given a described situation, how likely is each option? It runs Cloudflare's Clef-Flash decision model and returns a probability for every option with no text generation, through one tool called `advise`. The agent keeps its own reasoning and still decides; ember supplies the number. Examples on the site include whether a vague request such as fixing something is specific enough to start, whether a failing test is a flake, and whether a change to auth needs review.

The model can run on your own Apple Silicon Mac (32 GB of memory and an 18 GB one-time download), on a server you host such as a Mac or NVIDIA GPU machine shared by a team behind a token, or on a server someone else runs. The MCP process never loads the model; it starts the model server on the first call so the handshake stays instant.

The project reports testing on 538 situations: right 904 of 967 times when it was sure, and 81 percent of 1,565 questions overall. Typical answers take about 0.92 seconds on an M4 Max.

## Features

- **Probabilities, not prose:** Returns a calibrated probability for each option of a typed question.
- **Three setups:** Local Mac, a self-hosted server or a server run by someone else.
- **Question types:** Yes or no, pick one, or a 0 to 3 scale such as change risk.
- **Lazy model start:** The model server starts on first use, keeping the MCP handshake fast.
- **Published benchmarks:** A benchmark report shows accuracy and where it errs.
- **Teachable rules:** You can instruct your assistant to consult it at set points.

## Good for

- Decide whether a vague request needs a clarifying question
- Triage a failing test as flaky or real
- Gauge risk before pushing a change
- Share one inference server across a team

## Install

### Claude Code (Terminal)

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

### Claude Desktop (claude_desktop_config.json)

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

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

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

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

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

### Windsurf (~/.codeium/windsurf/mcp_config.json)

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

### Other clients (your client's MCP settings (most use this shape))

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

## Tools

- `advise`: Takes a state and typed questions and returns a calibrated probability for each option.

## FAQ

### What hardware does local mode need?

An Apple Silicon Mac with 32 GB of memory and about 18 GB of disk.

### Does my code leave the machine?

In local mode requests stay on the Mac; with a server they go only to the one you configure.

### Which agents does it support?

Claude Code, Codex, opencode and Kilo Code are named.

### Is it free?

Yes, it is free and open source.

### Is ember free?

Yes. ember is open source, so you can use and self-host it for free.

### How do I add ember to Claude Code?

Run this command in your terminal: claude mcp add ember --env EMBER_AUTH_TOKEN='<EMBER_AUTH_TOKEN>' -- uvx ember-advise

### What are the best alternatives to ember?

The closest alternatives to ember are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.

## Alternatives

- [REA](https://ai-powered.com/mcp/rea): REA is an open-source MCP server for developer tools.
- [Codebase Memory](https://ai-powered.com/mcp/codebase-memory): Codebase Memory is an open-source MCP server for knowledge and memory.
- [GitHub](https://ai-powered.com/mcp/github-2): GitHub is an open-source MCP server for developer tools.
- [Serena MCP: the IDE for your agent](https://ai-powered.com/mcp/serena-mcp-the-ide-for-your-agent): Serena MCP: the IDE for your agent is an open-source MCP server for developer tools.
- [XERJ](https://ai-powered.com/mcp/xerj): XERJ is an open-source MCP server for knowledge and memory.
- [Ouroboros](https://ai-powered.com/mcp/ouroboros): Ouroboros is an open-source MCP server for developer tools.
- [Claude Code Ultimate Guide](https://ai-powered.com/mcp/claude-code-ultimate-guide): Claude Code Ultimate Guide is an open-source MCP server for ai models and llm tooling.
- [Concord MCP](https://ai-powered.com/mcp/concord-mcp): Concord MCP is a freemium MCP server for developer tools.

---
Source: https://ai-powered.com/mcp/ember
