# Quickstart

Canonical source: https://maguyva.ai/docs/quickstart

License: text under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/); code samples under [MIT-0](https://opensource.org/license/mit-0); the Maguyva name, logos, brand assets and other code are not covered. Details: [maguyva.ai/license](https://maguyva.ai/license).

The fastest path from zero to a verified answer: get a key, index one repository you already know, connect your client, and ask one question you can already grade.

## 1. Get an API key

[Start free](https://app.maguyva.ai) — no credit card required. Your API key (it starts with `mgv_`) lives in [app.maguyva.ai](https://app.maguyva.ai).

## 2. Index a repository you already know

Connect a GitHub repository from the portal and let it index. Pick a small repo you can sanity-check fast — that makes it obvious whether Maguyva is grounding the model, instead of trying to index your whole company on day one.

## 3. Connect your client

For Claude Code, install the Maguyva plugin from the marketplace — run each command inside Claude Code:

```text
/plugin marketplace add maguyva/agent-plugins
/plugin install maguyva@maguyva
```

The plugin authenticates with your key from the `MAGUYVA_API_KEY` environment variable. Set it in your shell (per-project `direnv` is recommended, since keys are repository-scoped):

```bash
export MAGUYVA_API_KEY="mgv_xxxx"
```

Working on a team? Choose project scope in the panel that `/plugin install maguyva@maguyva` opens. That records the plugin in the repository's `.claude/settings.json`, which you commit; each teammate still runs the install once. The in-session `/plugin` commands take no `--scope` flag, so to script it, run the shell form from the repository root:

```bash
claude plugin marketplace add maguyva/agent-plugins --scope project
claude plugin install maguyva@maguyva --scope project
```

Most other clients (Cursor, VS Code, Devin Desktop, Zed, and friends) connect to the remote server directly — point them at the endpoint with a `Bearer` header, no bridge required. For example, Cursor's `.cursor/mcp.json`:

```json
{
  "mcpServers": {
    "maguyva": {
      "url": "https://maguyva.tools/mcp",
      "headers": {
        "Authorization": "Bearer ${env:MAGUYVA_API_KEY}"
      }
    }
  }
}
```

The shape varies slightly per client (VS Code nests under `servers` with `type: "http"`, Antigravity uses `serverUrl`, Zed uses `context_servers`). This Cursor example reads your key from the `MAGUYVA_API_KEY` environment variable shown above; where a client's config shows the `mgv_xxxx` placeholder instead, replace it with your API key. Then restart your client. The exact config for all 23 documented clients is in the [Installation Guide](https://maguyva.ai/docs/installation).

## 4. Ask one question you can already grade

Start with **"What repositories do I have connected?"** to confirm the connection, then ask one real question about your repo whose answer you already know — something like **"How is authentication handled in this codebase?"**

You should get back answers with file paths and line numbers. Open the cited files and check that the answer survives contact with the actual repo. That one graded answer tells you more than any feature list — make the tool earn trust before it joins your default loop.

No results or connection errors? Check your API key, client configuration, and repository indexing status in [Troubleshooting](https://maguyva.ai/docs/troubleshooting). If your setup uses the bridge, also check that `npx` is available.

## Next steps

- Installation Guide: per-client config for all 23 documented clients
- How It Works: what happens between your question and the answer
- MCP API Reference: every tool, parameter by parameter