# GitHub MCP server

> Wraps the gh CLI to give AI clients structured, token-efficient JSON for GitHub pull requests, issues, and Actions workflows.

- Repository: https://github.com/Dave-London/Pare
- Page: https://hostedmcp.org/servers/dave-london/github
- Transport: stdio
- Stars: 141
- License: MIT
- Latest version: 0.7.1

## Install

### Claude Code

```sh
claude mcp add github -- npx -y @paretools/github
```

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

```json
{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@paretools/github"
      ]
    }
  }
}
```

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

```json
{
  "servers": {
    "github": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@paretools/github"
      ]
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@paretools/github"
      ]
    }
  }
}
```

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

```toml
[mcp_servers.github]
command = "npx"
args = ["-y", "@paretools/github"]
```

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

```json
{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@paretools/github"
      ]
    }
  }
}
```

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `PARE_GIT_TOOLS` | no | no |  |
| `PARE_NPM_TOOLS` | no | no |  |
| `PARE_TOOLS` | no | no |  |

## FAQ

### How does the GitHub server reduce token usage?

Pare tools return structured, schema-validated JSON rather than raw CLI text, cutting tokens on verbose commands such as git log by around 92%.

### Can I limit which tools the GitHub server exposes?

Yes, set a per-server filter env var or the universal PARE_TOOLS variable to register only selected tools.

### How do I install the server?

Run the Pare init CLI to configure MCP servers non-interactively, then restart your client session.

### What output formats does a tool return?

Each tool returns human-readable content plus typed structuredContent validated against an output schema.