AI builders

yof in Claude Desktop

Paste one block into claude_desktop_config.json. A one-click bundle is built and waits to be published. Available, one detail to confirm.

Install

  1. In the yof dashboard, open AI builders (app.yof.io/yof/builders) and press Connect on the tool. yof makes a key for this connection and puts it inside the config, so you never type it. It is shown once.
  2. In Claude Desktop open Settings, Developer, Edit config, and paste the block.
  3. Add your project folder as the last argument, then restart Claude Desktop.

By hand, without a key (the account tools then explain how to connect):

{ "mcpServers": { "yof": { "command": "npx", "args": ["-y", "@yof/cli", "mcp", "/path/to/your/project"] } } }

Still to confirm: The one-click bundle (.mcpb) is built but not published yet; the config file works today.

What it can do

  • yof_check scans the project folder for leaked keys and made-up or risky packages.
  • yof_check_diff scans only what changed since the last commit, plus new files. Use it before you say a change is done.
  • yof_explain, yof_fix_prompt and yof_verify explain a finding, give the exact fix steps and re-scan the file.
  • With your yof key: yof_findings (your yof.io findings), yof_patch_status (Patch's pull requests and CI) and yof_lessons (notes from merged Patch fixes).

Privacy

  • The scans run on your computer. Your code and file contents are never sent anywhere.
  • The package check sends package names and pinned versions to npm, PyPI and OSV. Turn it off with offline: true.
  • With a yof key, the server sends one ping when it starts and, for the three account tools, only the tool name and its arguments (ids and rule ids) to mcp.yof.io.
  • Every answer is labelled as untrusted data from the scanned project, and secret values come back only as a short prefix plus REDACTED.