Small, sharp tools for AI agents and the people who run them. MIT where it can be, honest about its limits, and a first call with no signup wherever possible.
- agentguard — an MCP policy proxy: kill switch, hard
spend caps, destructive-action approvals, scoped per-agent keys, dry-run writes with a mutation
diff, a loop breaker, and a hash-chained audit log.
npx @agentwares/agentguard init. - overnight-kit — run a sequence of Claude Code prompts against a git repo while you sleep, with a digest in the morning.
- agentsmd-lint — lint AGENTS.md / CLAUDE.md / .cursorrules against the repo they actually describe.
- libs — the shared libraries behind the portfolio
(
@agentwares/mcp-kit,notify,x402,web-bot-auth). - servers — endpoints, pricing and tool references for the hosted services.
The hosted layer of agentguard (incidents, alerts, a tamper-evident audit log) has a free 1,000-call tier with no card: https://agentwares-agentguard.vercel.app/demo is a live dashboard.