PromptKing governs AI agents across six vendor surfaces using a single MCP endpoint. Paste the URL into any MCP-compatible client — your agent is governed in 60 seconds.
This page and the start_here tool teach the same protocol. Humans read the page; agents read the tool. Same truth.
MCP Endpoint
https://www.promptking32.com/api/mcp/mcpPaste into Claude Desktop, Cursor, or any MCP client. OAuth connects on first call.
System Context — Out-of-Path Architecture
PromptKing
↑ consumes metadata · emits decisions ↓
↓ Governance Decisions
↓ Evidence Receipts (SHA-256)
↓ MCP Governance Signals
↓ Your enforcement stack (SIEM · IT governance · vendor controls)
PromptKing does not touch agent traffic. Decisions travel — PromptKing does not.
Agent calls start_here. The port returns the governance contract: what tools exist, what the protocol expects, how to enroll. The agent learns the port without human configuration.
Agent registers via enroll or register_agent. Receives an agent ID, correlation ID, and autonomy tier. Identity is established — every subsequent call is attributed.
Before acting, the agent declares intent and cost via check_policy. The policy engine evaluates against org rules and returns a verdict: ALLOWED, APPROVAL_REQUIRED, or MODEL_RESTRICTION. No enforcement — the agent decides whether to proceed.
After acting, the agent reports what happened via report_outcome. An evidence receipt is issued with a SHA-256 integrity hash. Append-only — the receipt exists forever, independently verifiable at /verify/[id].
Governance Consultation
start_hereSelf-teaching protocol — agents learn the portenrollRegister an agent with identity + autonomy tiercheck_policyCHK-2: evaluate an action against org policyreport_outcomeCHK-3: declare outcome, receive evidence receiptIntelligence Readers
get_overviewOrg-level KPIs: spend, utilization, savingsget_seat_fleetPer-seat detail: plan, cost, recommendationget_seat_detailDeep dive on a single seatget_savings_opportunitiesRanked rightsizing recommendationsget_github_creditsGitHub AI Credits burn rate + model breakdownget_governance_scoresWatsonx model governance healthget_executive_reportCFO / COO / CRO persona reportsget_forecast30-day spend forecast with variance bandsGovernance Readers
get_trajectoryFull trajectory with linked governance artifactsget_trajectory_statusArchetype, cycles, cost, recent decisionsget_decision_receiptSingle policy decision as structured receiptlist_containment_signalsActive containment events for a trajectoryAgent Identity
register_agentCHK-1: register with SPIFFE identity + autonomy tierLive proof
Every governed run produces a receipt. Every receipt is independently verifiable — the hash is recomputed in your browser, not by our server.
Verify a real receipt →Protocol stats (live)
Loading live stats...
Numbers are live production counts — early-stage, honest, not inflated.
What we never touch
PromptKing processes metadata only — agent identity, cost declarations, policy verdicts, and governance signals. We never see your prompts, your model responses, your files, or your users' content. We never sit between your agent and the vendor API. Decisions travel. PromptKing does not.