agent-safety MCP servers & tools

MCP server for DashClaw governance — guard, record, invoke, and discover capabilities.

In-browser ~60Hz behavior tree execution engine with reactive triggers and safety guardrails.

Jev judgments for agents: injection scanning, shell risk gating, ranking

Agent 运行时全栈拦截: allow/deny tool calls, Ed25519 receipts, blocks RCE, theft, SSRF, injection. stdio.

Pre-flight spend authorization for AI agents. Set a budget, enforce limits, audit every decision.

Add human approval to AI agent workflows through cQnce authorization tools.

Ask a human before your agent acts: approval requests decided on a phone.

An effect gate for AI agents: ask before you act, so the same side effect happens at most once.

Local agent spend-policy checks. No funds, keys, payment authority, or network access.

Turn an IDF domain into an MCP server with invariants and role scopes in every tool description.

Answer CTRLRun approvals from an MCP client: list what is held, inspect it, grant or deny.

Foresee what git commands and multi-step changes will do, as calibrated probabilities

Agents can't skip steps: runs gated stepfiles on the client's own model.

Execution guard for AI agents — prevents duplicate tool calls on retry.

Gold-first AI-agent traceback diagnosis, exact fixes, verification, and reusable prescriptions.

Impact preview for AI agents - see what changes before any action executes.

Wraps the evolveguard CLI as a single generic MCP tool for agent-skill file safety checks.

Govern proposed Action Cards with Decision Receipts before developer-owned execution.

Anonymous public tools for FlightSweeper. See the published agent boundary before use.

Scan untrusted content for prompt injection and check link reputation before an agent acts

Human approval for AI agents: a named person approves or rejects, silence is not consent.

Pre-spend firewall for AI agents. Approves, blocks, flags transactions against policy rules.

Checks whether an agent's actions actually happened, against the destination's own records.

Local gut check before risky agent actions: calibrated safe/intended probabilities, no API calls.