Halto
MCP serverSafety gateway that checks MCP tool calls and holds risky ones for approval
Not checked yet- Remote
- Free
- Security
Safety gateway that checks MCP tool calls and holds risky ones for approval
Not checked yet
Auto-approve small agent actions under your rules; ask you about the rest
Not checked yetMCP server exposing a command approval gate and classify/route/score tools for coding
Runs locallyOpen-source gate that lets AI agents read freely but blocks unapproved writes to MCP
Runs locallyMCP operator server for answering held CTRLRun approvals from an AI assistant
Not checked yetPasskey-backed proof that a human approved a specific agent action
Not checked yetApproval gateway that holds AI-agent API writes for owner sign-off
Not checked yetExperimental MCP preflight check for consequential autonomous-agent actions before
Not checked yet
Human approval gate for AI agents, with a signed audit trail
Runs locallyA pre-action risk gate that scores irreversible agent actions GO, CAUTION or STOP
Runs locallyDeterministic preflight checks that block risky shell and email actions by AI agents
Not checked yetMCP server that pauses agent runs and routes approval requests to humans via Slack
Runs locallyMCP server for preflighting, approving and proving consequential AI agent actions
Not checked yetLocal signed receipts and approval gate for what AI agent sessions do
Runs locallyAdvisory governance checks for MCP hosts, with PII masking and audit traces
Runs locallyHuman approval receipts for AI agent actions, checkable offline by downstream services
Not checked yetPre-install trust check that lets agents assess MCPs, Skills and tools before enabling
Runs locally
Permission layer that holds consequential AI tool calls until a human approves the exact
Not checked yetOpen-source MCP server that blocks unattended destructive agent tool calls until a human
Runs locallyTrust and scam-risk scores that AI agents check before paying a counterparty
Not checked yetValidates and holds risky AI agent tool calls before they run, with human approval
Runs locallyThe most popular MCP servers for agent safety on AI-Powered right now are Halto, Approval Rules, Phocinae MCP, MCP Write Gate and CTRLRun Operator. Rankings combine upvotes, recent activity and, for open-source projects, GitHub stars.
Yes. Of the 21 MCP servers for agent safety listed, 10 are free or open source and 5 have a free plan. Use the Pricing filter to see only those.
AI-Powered lists 21 MCP servers for agent safety. New listings are added daily from the official MCP Registry, GitHub and maker submissions.