THE HUMAN GUIDE / PREVIEW

A little setup.
A different perspective.

MCP.pictures helps you discover the camera. Your camera’s local service is where you set it up and find its connection details.

Start with your local camera.

Connect to the same network as your camera, then open its local service in a new tab.

Open mcppics.local ↗

The local address works only when a camera or local service is running and your device can resolve and reach it. This website does not scan your network or redirect you automatically.

The product is in development. This guide does not claim that a device, firmware release, or working MCP endpoint is already available. The local service must provide the actual transport, endpoint, authentication, and supported tools.

Two addresses. Two different jobs.

Address	What it does
MCP.pictures	Product discovery and documentation.
mcppics.local	The camera’s local setup and service, reached from your own network.

Opening a local page in your browser does not give a cloud-based agent access to that network. The agent’s host must also reach the camera and resolve its name. A .local name is not a public internet endpoint.

Cloud clients cannot directly reach a .local address. Claude Desktop, Codex, and ChatGPT Desktop running on a computer on the camera’s network can connect locally using the instructions below. Do not publish the camera directly to the internet as a shortcut.

What setup will involve.

Place the camera where it can see your page or project.
Follow the device’s instructions to join your network.
Open the local service and get its MCP connection details.
Choose a compatible client and add the server using its supported setup flow.
Authorize the connection, inspect available tools, and request a test picture.

The device service must provide the confirmed MCP endpoint, transport, and any authentication details. The local client must run on a computer that can reach the camera.

Claude Desktop

Claude Desktop is supported when it runs on a computer that can reach the camera on the same local network. For an HTTP MCP service, Claude Desktop uses a local stdio bridge to connect to the camera.

Connect the computer running Claude Desktop to the camera’s network.
Open mcppics.local and get the MCP endpoint URL and transport from the device service. Use the endpoint it provides; do not assume the homepage or a guessed path is the MCP endpoint.
In Claude Desktop, open Settings → Developer → Edit Config. Add an entry inside the existing mcpServers object (merge it with any other entries):

{
  "mcpServers": {
    "mcppics": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "http://mcppics.local/<MCP_PATH_FROM_DEVICE>",
        "--allow-http"
      ]
    }
  }
}

Replace the example URL with the exact HTTP MCP endpoint shown by the device service. This bridge requires Node.js/npm so Claude Desktop can run npx. The HTTP option is intended only for a trusted local network. Save the config and restart Claude Desktop. Check Settings → Developer or the chat’s Connectors menu for the server and its tools.

Anthropic’s Claude Desktop MCP guide ↗ · mcp-remote bridge documentation ↗

The Claude Desktop client path is supported. The local service must provide a working MCP endpoint and transport.

Codex

Codex is supported when it runs locally on a computer that can reach the camera. Local Codex connects directly to a Streamable HTTP MCP endpoint; no cloud connector or bridge is needed.

Run Codex on the same network as the camera, then open mcppics.local and get the exact Streamable HTTP MCP endpoint from the device service.
In a terminal on that Codex host, add the endpoint:

codex mcp add mcppics --url "http://mcppics.local/<MCP_PATH_FROM_DEVICE>"

Replace the example URL with the exact endpoint from the device service. Alternatively, add it to ~/.codex/config.toml (or the project’s trusted .codex/config.toml):

[mcp_servers.mcppics]
url = "http://mcppics.local/<MCP_PATH_FROM_DEVICE>"

Verify with codex mcp list or /mcp in the Codex TUI. Codex’s desktop app, CLI, and IDE extension share the same host configuration. The Codex host must resolve and reach mcppics.local; a cloud Codex session cannot directly reach a private .local address.

OpenAI Docs: MCP in Codex ↗

The local Codex client path is supported. The device service must provide a compatible, working Streamable HTTP MCP endpoint.

ChatGPT Desktop

ChatGPT Desktop is supported for a local camera connection when the desktop app runs on a computer that can reach mcppics.local. Add the server in the app’s MCP settings; this uses the local app connection, not a cloud connector.

Run ChatGPT Desktop on the same network as the camera. Open mcppics.local and get the exact Streamable HTTP MCP endpoint from the device service.
In ChatGPT Desktop, open Settings → MCP servers → Add server. Enter a name, choose Streamable HTTP, and paste the exact endpoint URL supplied by the device.
Save the server and restart the app. In the composer, enter /mcp to confirm the server is connected and view its tools.

The ChatGPT desktop app shares local MCP configuration with Codex CLI and the IDE extension on that host. A ChatGPT web or cloud session cannot directly resolve or reach the private .local address.

OpenAI Docs: MCP in ChatGPT and Codex ↗

The ChatGPT Desktop local client path is supported. The device service must provide a compatible, working Streamable HTTP MCP endpoint.

What happens to the picture?

The camera supplies visual input; your connected agent interprets it. Once an image is shared with an AI service, that service’s data handling applies. Capture behavior, access controls, and retention must be documented by the device service before release.

For agents

Read the agent handoff guide ↗ · Markdown version ↗

A little setup.A different perspective.