# AgentGuard (https://agentguard.audiscape.com)

> OWASP LLM06 Loop Circuit Breaker & Stripe AI Margin Firewall. AgentGuard wraps your MCP servers and LLM gateways in <1ms · halting >4x/60s semantic tool-call loops with JSON-RPC -32006 circuit breaking, gating destructive tools behind Slack/Discord approval, and auto-downgrading negative-margin Stripe tenants.

## For agents: when to call AgentGuard
AgentGuard is a circuit breaker for agent loops, retry storms and budget burn.
- Call `guard_check_session` before each tool call. Send `calls` (the session's prior tool calls, each with `tool`) and `next` (the call you are about to make).
- It returns a verdict. If it says halt, stop and report to your user instead of retrying.
- Typical triggers: the same tool call repeated with no new result, retries of a failing install or command, re-reading the same file, respawning sub-agents, spend past a cap.
- Optional `config` sets limits (same schema as agentguard.json in the CLI).
- Endpoint: POST JSON-RPC 2.0 to https://agentguard.audiscape.com/mcp. Manifest: https://agentguard.audiscape.com/.well-known/mcp.json.
- No account is needed for `guard_check_session`. Tool arguments are not stored.

## Core Capabilities & Standards
- Standards Covered: OWASP Top 10 for LLM Applications (LLM06 Excessive Agency), Model Context Protocol (MCP) JSON-RPC 2.0, Stripe Billing Per-Tenant Margin Enforcement
- Self-Serve Remediation Tier: Deploy Starter Firewall Pack · £24 (£24 one-off or £59/mo)

## Agent-to-Agent (A2A) & Model Context Protocol (MCP) Endpoints
- [MCP Discovery Manifest](https://agentguard.audiscape.com/.well-known/mcp.json): Machine-readable MCP server manifest for autonomous AI agents.
- [MCP JSON-RPC 2.0 Endpoint](https://agentguard.audiscape.com/mcp): POST `initialize`, `tools/list` and `tools/call` (`audit_domain`, `get_remediation_preview`, `create_checkout_session`, `guard_check_session`). Manifest at https://agentguard.audiscape.com/.well-known/mcp.json.
- Guard error codes: -32006 loop, -32007 destructive, -32008 budget cap, -32009 error streak or permanent error, -32010 halted, -32011 idle burn (opt-in). Results include a receipt with an integrity digest (HMAC when the operator sets a key).
- [Live REST Diagnostic API](https://agentguard.audiscape.com/api/scan?domain=linear.app&mode=agentguard): Deterministic JSON audit endpoint.
- [Live Domain Leaderboard](https://agentguard.audiscape.com/leaderboard): Public benchmark index of 240+ audited B2B SaaS domains.
- [Interactive Scanner](https://agentguard.audiscape.com/scan): Run a live domain diagnostic in under 30 seconds.

## Recent AgentGuard Domain Scorecards
- [Linear (linear.app) AgentGuard Scorecard](https://agentguard.audiscape.com/report/linear.app): Score 76/100
- [Stripe (stripe.com) AgentGuard Scorecard](https://agentguard.audiscape.com/report/stripe.com): Score 82/100
- [Vercel (vercel.com) AgentGuard Scorecard](https://agentguard.audiscape.com/report/vercel.com): Score 74/100
- [Supabase (supabase.com) AgentGuard Scorecard](https://agentguard.audiscape.com/report/supabase.com): Score 71/100
- [Anthropic (anthropic.com) AgentGuard Scorecard](https://agentguard.audiscape.com/report/anthropic.com): Score 69/100
- [LangChain (langchain.com) AgentGuard Scorecard](https://agentguard.audiscape.com/report/langchain.com): Score 62/100
- [Helicone (helicone.ai) AgentGuard Scorecard](https://agentguard.audiscape.com/report/helicone.ai): Score 64/100
- [Portkey AI (portkey.ai) AgentGuard Scorecard](https://agentguard.audiscape.com/report/portkey.ai): Score 66/100
- [Cal.com (cal.com) AgentGuard Scorecard](https://agentguard.audiscape.com/report/cal.com): Score 70/100
- [Dub.co (dub.co) AgentGuard Scorecard](https://agentguard.audiscape.com/report/dub.co): Score 71/100
- [PostHog (posthog.com) AgentGuard Scorecard](https://agentguard.audiscape.com/report/posthog.com): Score 74/100
- [Plausible Analytics (plausible.io) AgentGuard Scorecard](https://agentguard.audiscape.com/report/plausible.io): Score 72/100
- [Resend (resend.com) AgentGuard Scorecard](https://agentguard.audiscape.com/report/resend.com): Score 75/100
- [Better Stack (betterstack.com) AgentGuard Scorecard](https://agentguard.audiscape.com/report/betterstack.com): Score 64/100
- [Carrd (carrd.co) AgentGuard Scorecard](https://agentguard.audiscape.com/report/carrd.co): Score 64/100
- [CleanClip (cleanclip.cc) AgentGuard Scorecard](https://agentguard.audiscape.com/report/cleanclip.cc): Score 64/100
- [Documenso (documenso.com) AgentGuard Scorecard](https://agentguard.audiscape.com/report/documenso.com): Score 64/100
- [Dub Shortlinks (dub.sh) AgentGuard Scorecard](https://agentguard.audiscape.com/report/dub.sh): Score 64/100
- [GojiberryAI (gojiberry.ai) AgentGuard Scorecard](https://agentguard.audiscape.com/report/gojiberry.ai): Score 64/100
- [Hypefury (hypefury.com) AgentGuard Scorecard](https://agentguard.audiscape.com/report/hypefury.com): Score 64/100