CACCHT ExplorerGuide← Back to the Explorer

Asking Claude or ChatGPT

You can ask Claude, in your own words, for a search: find all cases of >LHJM/ in the BHSA, or clauses where the object comes before the verb. Claude writes a search template, tries it, and sends it to the Explorer. You decide whether it runs. This uses your own Claude subscription, in Claude Desktop or Claude Code.

ChatGPT can do the same, in the ChatGPT desktop app and in Codex, with your own ChatGPT subscription; see In ChatGPT or Codex. What this chapter says about Claude holds for ChatGPT too.

What Claude does, and what it doesn't

Claude can be wrong. Check the template, and look at a few results before you trust a count.

Setting it up

Always first: start CACCHT Explorer and keep it running. Claude talks to the running program.

In Claude Desktop

  1. Download CACCHT-Explorer.mcpb from the release (under Assets, next to the program; see Getting started).
  2. Install it, in one of these ways:
    • double-click the downloaded file: Claude Desktop opens it and offers to install it; click Install;
    • or, in Claude Desktop, go to Settings → Extensions, and choose Install Extension… (under Advanced settings), then the file;
    • or drag the file into the Settings → Extensions window.
  3. Check that CACCHT Explorer is switched on under Settings → Extensions.
  4. Start a new chat. The tools menu (the button below the message box) now lists CACCHT Explorer.

You need a recent Claude Desktop. The extension needs nothing else: no Python and no Node.js of your own. If your Claude Desktop names its menus differently, look under Extensions for the button that installs one.

Without the extension (if installing it doesn't work): in Claude Desktop, Settings → Developer → Edit Config opens claude_desktop_config.json. Add the program, with its path on your computer, and restart Claude Desktop:

{
  "mcpServers": {
    "context-fabric": {
      "command": "C:\\path\\to\\cfabric-browser\\cfabric-browser.exe",
      "args": ["mcp"]
    }
  }
}

On a Mac the command is /Applications/CACCHT Explorer.app/Contents/MacOS/cfabric-browser; on Linux, the path of cfabric-browser/cfabric-browser.

In Claude Code

In a terminal, once:

claude mcp add --transport http context-fabric http://127.0.0.1:8107/mcp

8107 is the program's usual port. If another program had taken it, the log (.cfabric/logs/cfabric-browser.log in your home folder) gives the address with the port the Explorer uses.

In ChatGPT or Codex

New, and not yet tried with ChatGPT itself. Please tell us how it goes (see Questions and problems).

This works through Codex, OpenAI's assistant for working on a computer: in the ChatGPT desktop app, in Codex's own app, and in a terminal. The ChatGPT desktop app may offer it only when you work with Codex there, not in an ordinary chat. All of these share their settings: what you add in one, the others have too.

In the ChatGPT desktop app: open Settings → MCP servers and choose Add server. Give it the name caccht-explorer, choose Streamable HTTP, and enter the address

http://127.0.0.1:8107/mcp

Save it, then choose Restart. 8107 is the Explorer's usual port; see In Claude Code if it uses another.

In a terminal, once:

codex mcp add caccht-explorer --url http://127.0.0.1:8107/mcp

Both ways need the Explorer to be running when you ask.

Or let ChatGPT start the Explorer itself, as Claude Desktop does: add these lines to the file config.toml in the folder .codex in your home folder (make the file if it isn't there), with the program's path on your computer:

[mcp_servers.caccht-explorer]
command = "/Applications/CACCHT Explorer.app/Contents/MacOS/cfabric-browser"
args = ["mcp"]
tool_timeout_sec = 300

That is the path on a Mac. On Windows, write the path with single quotes, such as command = 'C:\path\to\cfabric-browser\cfabric-browser.exe'; on Linux, the path of cfabric-browser/cfabric-browser. On a Mac the folder .codex is hidden: in the Finder, choose Go → Go to Folder… and type ~/.codex. On Windows it is C:\Users\<your name>\.codex.

The first question can take longer than ChatGPT waits for a tool by default (a minute), because the Explorer may still have to open its corpora. If ChatGPT reports that the tool timed out, ask again. To make it wait longer, add the line tool_timeout_sec = 300 under [mcp_servers.caccht-explorer] in config.toml, as above.

ChatGPT on the web and on a phone can't reach a program on your computer, so none of this works there.

Asking

Ask Claude, for instance:

The first time Claude uses a tool, it asks your permission. Claude reads a guide to searching the BHSA, checks its template and tries it, and tells you it has sent a search.

In the Explorer, the search appears above the search form (in Parallel: click Search in the top bar), with your request, the template, Claude's explanation and how many results its trial found:

A search proposed by Claude, in Parallel

So far Claude knows the BHSA best: its guide to searching covers the BHSA's features, with worked examples. It can search the Samaritan Pentateuch and the scrolls too, but with less help.

Privacy

What you ask goes to Anthropic, as anything you ask Claude (or to OpenAI, when you ask ChatGPT). The texts stay on your computer, apart from the few results Claude sees when it tries a template. The Explorer answers only programs on your own computer, not other web pages or other computers.