Connect an AI assistant

MCP gives an assistant tools to work with your inventory. The endpoint is https://stuffowl.com/mcp, using Streamable HTTP.

Codex

Add the remote server and sign in through OAuth:

Terminal

codex mcp add stuffowl --url https://stuffowl.com/mcp
codex mcp login stuffowl

For a PAT connection, set your secret in the STUFFOWL_TOKEN environment variable and use this entry in your Codex configuration:

~/.codex/config.toml

[mcp_servers.stuffowl]
url = "https://stuffowl.com/mcp"
bearer_token_env_var = "STUFFOWL_TOKEN"

Restart or reload the client after changing its environment. Follow the Codex MCP guide for your client’s current configuration options.

Claude Code

Terminal

claude mcp add --transport http stuffowl https://stuffowl.com/mcp

Open Claude Code, run /mcp, choose StuffOwl, and complete authentication. For PAT setup, use an environment-variable reference in your MCP server configuration:

MCP configuration

{
  "mcpServers": {
    "stuffowl": {
      "type": "http",
      "url": "https://stuffowl.com/mcp",
      "headers": {
        "Authorization": "Bearer ${STUFFOWL_TOKEN}"
      }
    }
  }
}

See Claude Code’s MCP documentation for configuration scope and authentication options.

ChatGPT

Open ChatGPT’s app or connector settings. Enable developer mode if your account requires it for custom MCP apps.
Create a custom app named StuffOwl with the MCP URL above and OAuth authentication.
Connect, sign in to StuffOwl, and review the requested permissions.
Enable StuffOwl in the conversation where you want to use your inventory.

Custom MCP availability, write tools, and image-tool handling depend on the ChatGPT plan, workspace policy, and client. Follow the current ChatGPT connection guide. If your client does not display returned image content to the model, use a client with MCP image support.

Try a photo-naming workflow

Example request to your assistant

Find up to 10 unnamed items that have photos. Inspect each photo.
Suggest concise, useful names and descriptions. Do not guess brands
or model numbers. Create an enrichment preview and show me the
before-and-after changes before applying them.

list_inventory finds unnamed photographed items.
get_record or list_photos provides attachment IDs and context.
get_photo returns actual image content to the model.
preview_enrichment saves the proposed changes.
apply_enrichment applies a batch when authorized. undo_enrichment restores its fields if there have been no intervening edits.

For classification, ask the assistant to add descriptive tags. StuffOwl’s node kinds—place, box, bag, item—describe containment rather than object categories.

Available tools

Read inventory

list_inventory, get_record, search_inventory, list_kinds, get_history

Inspect images

list_photos, get_photo

Review suggestions

preview_enrichment, get_enrichment, list_enrichment, apply_enrichment, undo_enrichment

Organize

create_record, update_record, move_record, list_tags, add_tag, remove_tag

Export and labels

export_record, get_label, print_label, get_checks

Delete

delete_record (requires inventory:delete)

Tool discovery includes typed arguments and read/write annotations. Destructive operations still require their scopes. Tool errors appear as structured results with isError: true.

What this connection does—and what it costs

StuffOwl does not run a model inside the MCP server. Your assistant analyzes the returned images using its own model and usage allowance. Any additional AI fees depend on that provider and your plan.

A connected assistant acts when you ask it to. Automatic naming after every photo would require a separate background workflow. Recognition can be wrong; keep uncertain details out of names and review proposed changes.

The MCP server currently supports image inspection, but not binary uploads through tools. Use the REST API to upload photos and receipts. PDF receipts can be downloaded through the API; get_photo accepts images only.