# Live ToolMesh demo: connect Claude or ChatGPT in 30s

> Live ToolMesh demo at demo.toolmesh.io with two Hacker News backends — no API keys. Login dadl/toolmesh, then try 'Use ToolMesh to get the top 5 HN stories.'

Canonical: https://www.toolmesh.io/en/demo/

We run a public ToolMesh demo at **demo.toolmesh.io** — a single MCP endpoint, two HN backends, zero API keys. Just enough to experience the complete ToolMesh flow: OAuth login, tool discovery, credential injection, execution, and audit.

<Aside type="tip" title="Quick connect">
**MCP Endpoint:** `https://demo.toolmesh.io/mcp`
**Login:** `dadl` / `toolmesh`
</Aside>

## What to try

Once connected, try these prompts. Start with **"Use ToolMesh to"** to ensure your AI assistant uses the MCP tools instead of its built-in web search:

- **"Use ToolMesh to get the top 5 stories on Hacker News with their scores and comment counts."** — live data via the Firebase API
- **"Use ToolMesh to search HN for stories about 'credential isolation' — are there any recent discussions?"** — full-text search via Algolia, filtered by recency
- **"Use ToolMesh to compare the karma of HN users 'dang' and 'pg' — who has more?"** — two API calls, structured comparison

Try combining both backends in a single prompt:

> "Use ToolMesh to search HN for recent articles about AI agents, then get the full comment thread for the most discussed one."

:::tip
Some AI clients (especially ChatGPT) prefer their built-in web search over MCP tools for general questions. Prefixing your prompt with **"Use ToolMesh to"** tells the assistant to use the connected tools instead.
:::

## Connect

<Tabs>
<TabItem label="Claude Desktop">

### Via the GUI

1. Open a new chat and click the **+** button in the input area
2. Click **Connectors**
3. Click **Manage Connectors**
4. Click **+** (top right)
5. Click **Add custom connector**
6. Enter the URL: `https://demo.toolmesh.io/mcp`
7. Click **Add**

Claude Desktop opens a browser window for OAuth login. Log in with `dadl` / `toolmesh`.

{/* Screenshot: Claude Desktop connector setup flow */}

### Via config file (alternative)

Add to your config file (`~/Library/Application Support/Claude/claude_desktop_config.json` on macOS, `%APPDATA%\Claude\claude_desktop_config.json` on Windows):

```json
{
  "mcpServers": {
    "toolmesh-demo": {
      "url": "https://demo.toolmesh.io/mcp"
    }
  }
}
```

Restart Claude Desktop after editing.

</TabItem>
<TabItem label="Claude Code">

```bash
claude mcp add -t http -s user toolmesh-demo https://demo.toolmesh.io/mcp
```

Claude Code prompts you to authenticate via OAuth in your browser. Log in with `dadl` / `toolmesh`.

Verify the connection:

```bash
claude mcp list
```

</TabItem>
<TabItem label="ChatGPT">

Requires a **Plus, Pro, Business, Enterprise, or Education** plan.

1. Open **Settings > Apps & Connectors > Advanced settings** and toggle **Developer Mode** on
2. Go back to **Apps & Connectors**, click **Create**
3. Fill in:

| Field | Value |
|-------|-------|
| **Name** | ToolMesh Demo |
| **Description** | Live demo — Hacker News via ToolMesh |
| **MCP Server URL** | `https://demo.toolmesh.io/mcp` |
| **Authentication** | OAuth |

4. Check **I trust this application** and click **Create**
5. ChatGPT redirects to the ToolMesh login page — log in with `dadl` / `toolmesh`

{/* Screenshot: ChatGPT connector setup */}

</TabItem>
</Tabs>

## What's included

| Backend | Tools | What it does |
|---------|-------|-------------|
| **Hacker News** (Firebase API) | 10 | Browse stories, items, users, polls, and live feeds |
| **Algolia HN Search** | 4 | Full-text search across stories and comments |

The built-in **echo tools** are always available — useful for verifying your connection:

| Tool | What it does |
|------|-------------|
| `echo_echo(message)` | Returns whatever you send it |
| `echo_add(a, b)` | Adds two numbers |
| `echo_time()` | Returns the current UTC time |

All tool calls run through the full ToolMesh pipeline: authentication, credential injection, execution, and audit logging.

## About this demo

This demo is intentionally narrow: one public MCP endpoint, two read-only HN backends, no custom APIs — just enough to experience the complete ToolMesh flow without any setup overhead. It is a shared instance with no SLA.

The [DADL registry](https://dadl.ai) ships definitions for a growing catalog of public APIs (live count on [dadl.ai](https://dadl.ai)). To connect your own APIs, run your own instance:

- [Getting Started](/en/getting-started/) — self-hosted setup walkthrough
- [Connect Claude](/en/setup-claude/) — detailed setup for all Claude products
- [Connect ChatGPT](/en/setup-openai/) — ChatGPT Developer Mode setup