agentguard
MCP policy proxy: spend caps, approvals for destructive tools, kill switch, dry-run, audit log.
Open source Open in the app JSON README (API)
About
MCP policy proxy: spend caps, approvals for destructive tools, kill switch, dry-run, audit log.
Details
- Kind
- MCP servers
- Topic
- Security & identity
- Publisher
- agentwares
- Origin
- official
- Category
- ferramentas
- Transport
- local
- Version
- 0.1.3
- Last push
- 2026-09-07T20:10:59Z
- Repository state
- ativo
- Language
- TypeScript
- License
- MIT
- Added
- 2026-09-07 16:05:13
- Updated
- 2026-09-08 01:01:06
- Origin id
io.github.agentwares/agentguard
README
# agentguard A policy proxy between an AI agent and the tools it can reach: hard spend limits, approval before anything destructive, a kill switch, per-agent scoped credentials, dry-run writes with a mutation diff, a semantic loop breaker, and a hash-chained audit log you can verify. ```bash npx @agentwares/agentguard init ``` No LLM calls anywhere in it. No phone-home. No account. MIT. - `packages/agentguard-core` — the policy engine (Web-standard, runs anywhere) - `apps/agentguard-cli` — `agentguard`, the proxy and its commands - `packages/agentguard-sdk` — the same engine for non-MCP tool calls - `assets/permission-diff-action` — a GitHub Action that comments when agent permissions widen - `packages/mcp-kit`, `packages/notify`, `packages/x402` — generic plumbing the CLI bundles Hosted incident dashboard, audit retention and alerting: https://agentwares-agentguard.vercel.app This repository is exported from a private monorepo, which is why it has no long history. The exported source is complete and is exactly what is published to npm.