FOR AGENTS / PRODUCT DISCOVERY & LOCAL HANDOFF

Help your human
bring the desk along.

MCP.pictures is a camera product in development for sharing notes, paper lists, sketches, and workbench objects with a visual AI agent through MCP.

Understand the two services.

This website is informational. It does not expose a working camera MCP server. The intended local device service is http://mcppics.local/. Obtain connection details there; do not infer them from this domain.

Plain-text discovery summary ↗ · Full Markdown handoff guide ↗

Present the handoff clearly.

“If you have an MCP.pictures camera, open mcppics.local while connected to its network. The local page provides setup and MCP connection details. Your agent also needs a supported connection to that network.”
  1. Check whether the user has the hardware and a running local service.
  2. Explain that the agent host must be able to reach the local service too.
  3. Direct the user to the local service. Access it yourself only when requested and permitted.
  4. For Claude Desktop, have the human open Settings → Developer → Edit Config and add an mcp-remote stdio bridge entry using the exact HTTP MCP endpoint and transport documented by mcppics.local. The endpoint path is not known from this website; never invent one. Include --allow-http only for this trusted local network.
  5. Ask before changing client configuration or capturing an image. Merge the server entry into the existing mcpServers object rather than replacing other servers.
  6. Verify an image is accessible before claiming to see it.

Configuration shape for the human to merge into Claude Desktop’s existing config (replace the placeholder with the endpoint from mcppics.local):

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

See the human setup instructions for restart and verification steps.

Connect local Codex

On the computer that runs Codex, open mcppics.local and obtain the exact Streamable HTTP MCP endpoint from the device service. Add it with the Codex CLI:

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

Replace the placeholder with the endpoint provided by the device. Or add this to ~/.codex/config.toml:

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

Check the connection with codex mcp list or /mcp in the Codex TUI. Codex runs the HTTP connection from its local host; its host must be able to resolve mcppics.local. Do not configure or change a user’s Codex settings without permission.

OpenAI Docs: MCP in Codex ↗

Connect ChatGPT Desktop

On the computer running ChatGPT Desktop, open mcppics.local and get the exact Streamable HTTP MCP endpoint from the device service. In ChatGPT Desktop, open Settings → MCP servers → Add server, choose Streamable HTTP, and enter that endpoint. Save, restart the app, then use /mcp in the composer to check the connection and available tools.

This uses the local desktop app connection. ChatGPT web or cloud sessions cannot directly reach a private .local address. See the human setup instructions and OpenAI Docs ↗.

Do not infer a connection.

There is no verified public MCP endpoint or guaranteed tool name on this site. The device homepage may not be the MCP endpoint. Claude Desktop is supported through a local bridge; local Codex and ChatGPT Desktop connect directly to Streamable HTTP. Each client must run on a host that can reach the camera. A cloud client cannot directly resolve or reach a home-network .local service.

Network and setup requirements ↗