"author:ubukhari" matching MCP servers:

An MCP policy proxy that sits between AI agents and their upstream MCP servers to enforce hard spend and blast-radius caps, human approvals for destructive tools, a kill switch, scoped per-agent credentials, dry-run writes with mutation diffs, semantic loop breaking, and a hash-chained audit log. It also offers hook mode that applies the same policy to coding agents' own shell commands and file edits.

Reads a GitHub repository's own workflow and scheduled-run history to report runs that were late, never happened, failed, or succeeded without doing anything, plus schedules GitHub has disabled and the 60-day inactivity countdown. It is read-only, uses the caller's own GitHub token, and returns a JSON report usable by an agent to verify that scheduled workflows actually ran when they should have.

Enables checking a remote MCP server's OAuth sign-in chain from outside, identifying the first broken step, whose responsibility it is, and the fix, with client-specific verdicts for claude.ai, Claude Code, ChatGPT, VS Code, and Cursor.

A local, read-only MCP server that reconciles Stripe payouts against bank deposits and QuickBooks or Xero books against bank statements month by month from existing CSV exports, reporting what does not tie with the likely reason. It accepts file paths only, requires no account or bank linking, and writes nothing.