# Predictive Debugger MCP server

> Predicts runtime failures in JavaScript and TypeScript code to help coding agents identify risky files and debug potential errors.

- Repository: https://github.com/SpeedosDK/predictive-debugger
- Page: https://hostedmcp.org/servers/speedosdk/predictive-debugger
- Transport: stdio
- Stars: 2
- License: MIT
- Latest version: 0.9.0

## Install

### Claude Code

```sh
claude mcp add predictive-debugger -- npx -y predictive-debugger
```

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

```json
{
  "mcpServers": {
    "predictive-debugger": {
      "command": "npx",
      "args": [
        "-y",
        "predictive-debugger"
      ]
    }
  }
}
```

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

```json
{
  "servers": {
    "predictive-debugger": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "predictive-debugger"
      ]
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "predictive-debugger": {
      "command": "npx",
      "args": [
        "-y",
        "predictive-debugger"
      ]
    }
  }
}
```

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

```toml
[mcp_servers.predictive-debugger]
command = "npx"
args = ["-y", "predictive-debugger"]
```

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

```json
{
  "mcpServers": {
    "predictive-debugger": {
      "command": "npx",
      "args": [
        "-y",
        "predictive-debugger"
      ]
    }
  }
}
```

## Tools (7)

- `scan_project`
- `analyze_file`
- `check_types`
- `map_dependencies`
- `analyze_logs`
- `predict_failures`
- `list_providers`

## FAQ

### Do I need a separate API key for model predictions?

No, the server uses the model access and usage allowance of your already installed CLI, such as Claude Code, Codex, or GitHub Copilot CLI.

### What programming languages are supported?

The server supports JavaScript and TypeScript, including JSX, TSX, and decorators, though Vue and Svelte single-file components are not supported.

### How does the server handle privacy and data?

Static analysis, dependency maps, and log analysis are performed locally. Only the predict_failures tool sends source and dependency context to your CLI's model provider.