Constitutional policy governance for WordPress MCP. Every AI agent action passes through a deterministic policy gate before touching your site. The AI doesn't decide. The gate enforces.
// AI agent attempts bulk deletion
WPG_PolicyGate::evaluate([
'action' => 'core/delete-posts',
'args' => ['count' => 847],
'agent' => 'claude-desktop',
]);
// ✗ DENIED — Policy: no-bulk-delete
// "Bulk deletion of 10+ items
// requires human approval"
//
// ✓ Logged to immutable audit trail
// ✓ Agent notified with reason
// ✓ Zero items deletedWordPress's MCP Adapter lets AI agents create posts, delete users, modify settings,
and execute arbitrary abilities on your site. The only protection? A permission_callback that
checks WordPress roles. That's not governance. That's a checkbox.
WordPress checks if an agent can act, not whether it should. An admin-level AI agent can delete your entire site with a single MCP call.
Monitoring tools can show you what MCP requests came in — after the fact. But watching an agent delete 500 posts and having a log entry about it is not governance. You need a gate, not a dashcam.
There's no concept of "this agent should never do X regardless of permissions." WordPress trusts the role. AOS trusts the rules.
AOS Governance for WP is a drop-in plugin that wraps every MCP Ability in a constitutional policy gate. Install, activate, govern.
Every AI agent action is evaluated against your policy rules before execution. No LLM in the governance loop — pure deterministic logic. Sub-millisecond decisions.
CoreEvery action — approved or denied — is logged with full context: agent identity, action attempted, policy matched, timestamp, and outcome. Exportable and tamper-evident.
CoreShip with battle-tested policies: no bulk deletion, no admin creation, no plugin installation, content moderation gates, and more. Activate with one click.
CoreCreate custom governance rules without writing code. Drag-and-drop conditions, actions, and outcomes. Export as JSON or PHP for version control.
ProInstant webhook, email, or Slack notifications when a policy blocks an AI action. Know immediately when an agent tries something it shouldn't.
ProAssign different policy profiles to different AI agents. Claude gets content creation privileges. GPT gets read-only. Custom agents get custom rules.
ProManage governance policies across all your WordPress sites from a single centralized dashboard. Push policy updates to 100 sites simultaneously.
AgencyGenerate audit reports for HIPAA, SOC2, GDPR, and custom compliance frameworks. Prove your AI agents operated within policy at all times.
AgencyRegisters its own WordPress Abilities so AI agents can query governance state: check policies, read audit logs, and understand their boundaries natively via MCP.
CoreWordPress already has audit logs. MCP now has request trackers. But none of them can stop an AI agent from acting. That's the gap AOS closes.
Other tools answer: "What did the agent do?" — AOS answers: "Should the agent be allowed to do it?"
AOS injects a constitutional layer between WordPress's MCP Adapter and Ability execution. Zero modifications to core. Zero modifications to your plugins.
Claude, Cursor, or any MCP client calls a WordPress Ability (e.g., core/delete-posts)
Standard permission_callback verifies the user role has the capability. This is where it
normally stops.
Deterministic evaluation against your constitutional rules. Action, arguments, agent identity, and context are all assessed. No LLM. No ambiguity.
Approved actions proceed normally. Denied actions return a structured error with the policy reason. Everything is logged immutably.
// AI agent with admin role:
Agent: "Delete all posts tagged 'draft'"
WordPress: "Is user admin? Yes."
Result: 847 posts deleted ✗
Audit: Nothing logged ✗
Undo: Not possible ✗// AI agent with admin role:
Agent: "Delete all posts tagged 'draft'"
WordPress: "Is user admin? Yes."
WPG Gate: "Policy no-bulk-delete triggered"
→ DENIED: count 847 exceeds limit 10
→ Logged with full context ✓
→ Alert sent to admin ✓
→ Zero posts deleted ✓wp_ability_pre_execute filterwpg/check-policyToolPre-flight check if an action is allowed by policywpg/audit-logResourceQuery the immutable audit trail with filterswpg/active-policiesResourceList all active governance policies and ruleswpg/agent-profileResourceView the current agent's permissions and
restrictionswpg/report-generateToolGenerate compliance report for a date rangeEvery tier includes the deterministic policy engine. Upgrade for advanced features, multi-agent support, and cross-site governance.
Essential governance for personal sites
Full governance for a single site
14-day money-back guarantee
Govern AI across 10 sites
14-day money-back guarantee
No. AOS Governance for WP works on top of WordPress's existing role/capability system. WordPress checks if a user can do something. AOS checks if they should. Both must pass for an action to execute.
Absolutely not. AOS is 100% deterministic. Policies are pattern-matched rules evaluated in sub-millisecond time. No AI model is consulted during governance — that would defeat the purpose. The gate enforces. It doesn't think.
The agent receives a structured error with the exact policy name and reason. Your audit log records the denial. You can then adjust your policy rules through the visual builder or config files and the action will succeed on retry.
No. Policy evaluation adds less than 1ms per MCP call. AOS makes zero external API calls, zero LLM inferences, and uses a single lightweight database table. It's lighter than most analytics plugins.
AOS works with or without the MCP Adapter. If you have the Adapter installed, it automatically protects all Abilities exposed via MCP. Without it, AOS still governs any direct Ability calls via PHP or the REST API.
We're offering a one-time payment option that gives you permanent access to Pro or Agency features with no recurring fees. This is limited to the first 500 licenses. Once they're gone, we switch to annual-only pricing.
Yes. AOS governance technology is protected by patent-pending filings covering deterministic policy gate architecture, constitutional AI enforcement, and agent governance infrastructure. View our evidence repository →
Yes. AOS Governance for WP is transport-agnostic. It works with STDIO connections (local via WP-CLI) and HTTP connections (remote via the Automattic proxy). Any MCP-compliant client is supported.
MCP Tracker and activity logs are monitoring tools — they record what happened after the fact. AOS Governance for WP is a governance layer — it evaluates policy before execution and blocks actions that violate your rules. Think of it this way: MCP Tracker is your security camera. AOS is your security guard. You want both, but only one can actually stop the intruder.
Install in 60 seconds. Start governing immediately. No configuration required for the default policy set.
Free forever for personal sites. Pro and Agency tiers for professionals.