# Connect the FlyMyAI cloud to your AI client

> Add the entire FlyMyAI cloud - all models, your agents and 1,500+ tools - to Claude, Cursor, VS Code, ChatGPT or Codex over MCP. One-click OAuth, no API key to paste.

The **entire FlyMyAI cloud** is available over **MCP** - every AI model
(image, video, audio, music, speech, LLMs), your **autonomous agents**, and
**1,500+ tools** (Slack, Notion, GitHub, Google, CRMs, web search
and more). Add it to any MCP-capable client and authorize with your FlyMyAI
account right in the browser - no API key to copy or store.

**MCP URL** (the same everywhere):

```
https://mcp-agents.flymy.ai/mcp
```

## Pick your client

- [Claude Desktop](#claude-desktop--claudeai): OAuth · custom connector
- [claude.ai](#claude-desktop--claudeai): OAuth · custom connector
- [Claude Code](#claude-code--cli): OAuth or API key
- [Cursor](#cursor--vs-code): OAuth · remote MCP
- [VS Code](#cursor--vs-code): OAuth · remote MCP
- [ChatGPT](#chatgpt--openai-codex): OAuth · dev mode
- [OpenAI Codex](#chatgpt--openai-codex): OAuth · remote MCP
- [OpenClaw](#openclaw--hermes--other-agents): OAuth · remote MCP
- [Hermes](#openclaw--hermes--other-agents): remote MCP

## Claude Desktop / claude.ai

1. **Settings → Connectors → Add custom connector**.
2. Paste the **MCP URL** above and give it a name.
3. A browser window opens - **sign in to FlyMyAI** and press **Allow** on the
   consent screen.
4. The connector shows **Connected**. Ask Claude to run your agents, generate
   media, or use any tool.

## Claude Code (CLI)

OAuth (no key - the CLI registers and opens the browser for you):

```bash
claude mcp add --transport http flymyai https://mcp-agents.flymy.ai/mcp
```

Prefer your API key instead of OAuth? Pass it as a header:

```bash
claude mcp add --transport http flymyai \
  "https://mcp-agents.flymy.ai/mcp" \
  --header "X-API-Key: <YOUR_FLYMYAI_API_KEY>"
```

## Cursor / VS Code

Add a **custom / remote MCP server** pointing at the MCP URL. The editor opens
the FlyMyAI login + consent in your browser, then connects. Both register
automatically - no manual client setup.

## ChatGPT / OpenAI Codex

In **ChatGPT**, custom MCP connectors live behind **Developer Mode** (beta;
Plus / Pro / Team / Enterprise — on managed workspaces an admin enables it
first):

1. **Settings → Connectors → Advanced → Developer mode** → ON.
2. **Create** a connector → **MCP Server URL** = the URL above →
   **Authentication: OAuth** → confirm the trust checkbox → **Create**.
3. Finish the FlyMyAI **OAuth** login + consent in the browser, then enable it
   in a chat via **＋ → Developer mode → FlyMyAI**.

ChatGPT's connector UI is **OAuth-only** — there's no field to paste an API key.
**OpenAI Codex** (CLI) adds it as a remote MCP server and authorizes the same way.

## OpenClaw / Hermes / other agents

Most MCP-capable agents take the same URL as a **remote / streamable-HTTP MCP
server** and run the OAuth login in your browser:

- **OpenClaw** — add it and log in:

  ```bash
  openclaw mcp add flymyai --url https://mcp-agents.flymy.ai/mcp \
    --transport streamable-http --auth oauth
  openclaw mcp login flymyai
  ```

  (Header mode instead: `openclaw mcp set flymyai '{"url":"…","transport":"streamable-http","headers":{"X-API-Key":"YOUR_KEY"}}'`.)
- **Hermes Agent** (Nous Research) — add the URL as an MCP server in its config
  and authorize.

## Always allow FlyMyAI (skip the per-call prompt)

By default some clients confirm every tool call. Here's how to let FlyMyAI run
without prompting — do this only for tools you trust to run unsupervised.

### Claude Desktop / claude.ai

When Claude first calls a FlyMyAI tool, the approval dialog shows **Allow once**
and **Allow always** — click **Allow always**. It's per-tool, so the first use of
each tool prompts once, then never again. *Team / Enterprise:* an Owner can
pre-set this in **Settings → Connectors → FlyMyAI → Tool permissions** (set tools
to **Always allow**).

### Claude Code (CLI)

Add a server-wide allow rule to your settings — `~/.claude/settings.json` (all
projects) or `.claude/settings.json` (this project):

```json
{ "permissions": { "allow": ["mcp__flymyai"] } }
```

`mcp__flymyai` auto-approves every tool from the server you named `flymyai` (use
the name you passed to `claude mcp add`). For a single run instead:
`claude --allowedTools "mcp__flymyai__*"`. You can also pick **"Yes, and don't
ask again"** at the first prompt.

### OpenClaw

OpenClaw asks through its `permissions_respond` prompt with **allow-once /
allow-always / deny** — pick **allow-always** the first time each FlyMyAI tool
runs. (There's no documented config-file blanket auto-approve; `toolFilter` only
narrows which tools are exposed, it doesn't pre-approve them.)

### Hermes Agent

Hermes gates only risky shell commands, **not** MCP tools — FlyMyAI tools run
without a per-call prompt by default. To silence all approval prompts globally,
set `approvals.mode: off` in `~/.hermes/config.yaml` (or run `hermes --yolo`).

### ChatGPT

ChatGPT has **no persistent always-allow** for connectors. When the confirmation
appears, approve it and tick **remember** — but that lasts only for the current
conversation; new chats ask again. (Read-only tools generally run without a
prompt.)

## What you're granting

The consent screen names the app requesting access and what it can do - run your
agents, models and tools on your behalf. Access tokens are **scoped,
short-lived and revocable**, and FlyMyAI **never hands your API key to the
client**. Revoke access anytime by removing the connector.

> **Tip: Both methods work**
>
> OAuth and the `X-API-Key` header reach the **same** cloud MCP and run under your
> account. Use whichever your client supports.

Get an API key from [flymy.ai](https://flymy.ai).
