# understand-quickly MCP server

> Exposes a registry of code-knowledge graphs to AI clients for repository analysis, concept searching, and drift detection.

- Repository: https://github.com/looptech-ai/understand-quickly
- Page: https://hostedmcp.org/servers/looptech-ai/understand-quickly
- Transport: stdio
- Stars: 48
- License: Apache-2.0
- Latest version: 0.1.1

## Install

### Claude Code

```sh
claude mcp add understand-quickly -- npx -y @looptech-ai/understand-quickly-mcp
```

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

```json
{
  "mcpServers": {
    "understand-quickly": {
      "command": "npx",
      "args": [
        "-y",
        "@looptech-ai/understand-quickly-mcp"
      ]
    }
  }
}
```

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

```json
{
  "servers": {
    "understand-quickly": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@looptech-ai/understand-quickly-mcp"
      ]
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "understand-quickly": {
      "command": "npx",
      "args": [
        "-y",
        "@looptech-ai/understand-quickly-mcp"
      ]
    }
  }
}
```

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

```toml
[mcp_servers.understand-quickly]
command = "npx"
args = ["-y", "@looptech-ai/understand-quickly-mcp"]
```

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

```json
{
  "mcpServers": {
    "understand-quickly": {
      "command": "npx",
      "args": [
        "-y",
        "@looptech-ai/understand-quickly-mcp"
      ]
    }
  }
}
```

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `UNDERSTAND_QUICKLY_REGISTRY` | no | no |  |

## FAQ

### What tools does the server provide?

The server exposes four tools: list_repos, find_graph_for_repo, get_graph, and search_concepts.

### Does this server support authentication?

No, the current version is a minimal implementation with no authentication.

### How does the search_concepts tool work?

It primarily uses a precomputed stats.json file for aggregated concept matches, falling back to a cross-graph scan if the stats file is unavailable.