Provenancewhere it came from, and what was verified
Published by the owner of tooloracle.io, who proved control of the domain through the official MCP registry.
🛡 Permissions & trustwhat this server can touch before you run it
About
AgentGuard — 20-tool AI safety MCP: policy preflight, risk scoring, audit logging, rate limits.
Packages & transports
Metadata
Computed from the server's own data, not a hidden formula. This measures how much you can depend on it, not how popular it is. Stars are shown separately, on purpose. How it's scored →
▸policy_preflight100A
▸tool_risk_score100A
▸approval_required100A
▸audit_log_write100A
▸audit_log_query100A
▸decision_explain100A
▸rate_limit_check100A
▸payment_policy_check100A
▸spend_limit_check100A
▸secret_exposure_check100A
▸payload_safety_check92A
▸replay_guard_check100A
▸cross_tool_anomaly_check100A
▸scope_check75B
▸session_validate75B
▸tenant_policy_check75B
▸threat_intel_check83B
▸output_safety_scan94A
▸emergency_kill93A
▸tool_manifest_verify93A
▸approval_resolve100A
▸approval_list93A
▸policy_register93A
▸guard_metrics93A
Structural, not an LLM's opinion: each check is a fact about the tool's schema, so it's reproducible. One weak tool drags the score (60% average, 40% worst). How it's scored →
Raise this grade
up to +25 on qualityOwn this server? Here's exactly what to change, biggest wins first. Every number below is a fixed point value, so you can see what each edit is worth before you make it.
Push to the repo or cut a release. Activity within the last 30 days earns the full 45.
With no reviews this sits at a neutral 11. Reviews reporting it works push it toward the full 20.
Add an SPDX license (like MIT or Apache-2.0) to the repo so it's detected.
- 5 of 5 params have no description: `role`, `scopes`, `agent_id`, `tool_name`, +1 more. Describe each one.
- 8 of 8 params have no description: `role`, `action`, `scopes`, `agent_id`, +4 more. Describe each one.
- 4 of 4 params have no description: `action`, `agent_id`, `tenant_id`, `tool_name`. Describe each one.
- 2 of 3 params have no description: `agent_id`, `entity_type`. Describe each one.
Already earned (3) ▾Already earned (3) ▴
These are structural checks, so the fixes are exact. See the full formula →
Add a badge to your README
[](https://socketcat.com/servers/io.tooloracle/agentguard)Shows your live quality grade and links back here. It updates itself as the grade changes.
Reviews
Sign in to leave a reviewNo reviews yet. Be the first to say whether it worked.
{
"mcpServers": {
"agentguard": {
"command": "npx",
"args": [
"-y",
"agentguard"
]
}
}
}▶ Try itlive against feedoracle.io
This server has a hosted endpoint, so we can connect and list its actual tools right here. Nothing runs on your machine.