# Ekbasis MCP server

> An open consequence model that predicts the outcomes of git commands and multi-step actions with calibrated probabilities.

- Repository: https://github.com/OpenInterpretability/ekbasis
- Page: https://hostedmcp.org/servers/openinterpretability/ekbasis
- Transport: stdio
- Stars: 18
- License: Apache-2.0
- Latest version: 0.1.42

## Install

### Claude Code

```sh
claude mcp add --env EKBASIS_API_KEY=<YOUR_EKBASIS_API_KEY> ekbasis -- uvx ekbasis
```

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

```json
{
  "mcpServers": {
    "ekbasis": {
      "command": "uvx",
      "args": [
        "ekbasis"
      ],
      "env": {
        "EKBASIS_API_KEY": "<YOUR_EKBASIS_API_KEY>"
      }
    }
  }
}
```

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

```json
{
  "servers": {
    "ekbasis": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "ekbasis"
      ],
      "env": {
        "EKBASIS_API_KEY": "<YOUR_EKBASIS_API_KEY>"
      }
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "ekbasis": {
      "command": "uvx",
      "args": [
        "ekbasis"
      ],
      "env": {
        "EKBASIS_API_KEY": "<YOUR_EKBASIS_API_KEY>"
      }
    }
  }
}
```

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

```toml
[mcp_servers.ekbasis]
command = "uvx"
args = ["ekbasis"]

[mcp_servers.ekbasis.env]
EKBASIS_API_KEY = "<YOUR_EKBASIS_API_KEY>"
```

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

```json
{
  "mcpServers": {
    "ekbasis": {
      "command": "uvx",
      "args": [
        "ekbasis"
      ],
      "env": {
        "EKBASIS_API_KEY": "<YOUR_EKBASIS_API_KEY>"
      }
    }
  }
}
```

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `EKBASIS_API_KEY` | no | yes | Key for the hosted API (ekb_..., from https://openinterp.org/console); not needed for your own server |
| `EKBASIS_URL` | no | no | Ekbasis server: the hosted API, or your own server (open weights) |

## FAQ

### What does Ekbasis do?

It is a consequence model for agents that answers questions about what will happen after an action, such as whether a command will lose work or fail, using calibrated probabilities.

### How is Ekbasis different from a standard LLM?

Unlike LLMs that generate text through reasoning, Ekbasis is a world model that provides calibrated distributions over the next state's variables in a single forward pass.

### Can I use Ekbasis to protect my git repositories?

Yes, it can act as a git guard for coding agents by reading the repository state and checking commands before they are executed.

### What information is sent to the server?

The server receives the state of each check as text, including the command line (with redacted credentials) and a listing of what the line could touch, such as file names and sizes.