One URL to Governed

What Just Changed

Until today, connecting an AI agent to a governance platform required configuration files, developer tools, API keys generated through database queries, and a working knowledge of transport authentication. The distance between 'I want governance' and 'my agent is governed' was measured in hours and engineering tickets. That distance just collapsed to one URL. Paste it into claude.ai. Click Connect. Approve. Your agent discovers the governance protocol on its own and completes a four-stage handshake — enrollment, registration, policy check, outcome receipt — without anyone explaining the steps. The first time we ran it, the agent finished in under 60 seconds. Nobody from PromptKing was in the room.

How It Works

When you paste the PromptKing URL into any MCP-compatible client, the client sends an unauthenticated request. PromptKing responds with a standard discovery chain that tells the client exactly how to authenticate: where to register, where to get authorization, and where to exchange credentials. The client handles it automatically — the user sees a consent screen, approves, and lands in a governed session with 18 tools available. The consent screen is honest about what it grants: read access to spend, seat, and governance data, plus the ability to run governed agent workflows. It explicitly states what it does not grant — no recommendation acceptance, no settings changes. Every sentence is verifiable against the actual tool surface.

The Self-Teaching Protocol

The agent does not need documentation. The first tool it calls returns the complete governance protocol: four stages, required fields per stage, concrete examples, the full error taxonomy with repair instructions. A cold agent — one that has never seen PromptKing before — discovers the protocol, enrolls, registers itself with a human principal, requests a policy decision, and records an outcome receipt. The receipt includes a SHA-256 hash, a correlation identifier, and a cost variance against declared estimates. That receipt is the proof the run was governed.

What This Means for Enterprise AI

The barrier to AI governance just dropped from 'schedule an implementation' to 'paste a URL.' An IT Director evaluating governance tooling can connect, run a governed session, and see real policy decisions in their first sitting — before any procurement conversation. A developer building an agentic workflow can add governance in the time it takes to paste one line. The governed agent handshake works the same way whether you connect through claude.ai, Claude Desktop, or any MCP-compatible client. One URL. One protocol. Every agent.

The Out-of-Path Principle

PromptKing does not sit in the execution path. It does not see your prompts, read your data, or control your agent. It receives declared intents, evaluates them against policy, and issues receipts. Your stack enforces. PromptKing decides. The receipt proves it. The consent screen reflects this: governance access means your agent can enroll, request policy decisions, and record outcomes. It cannot accept recommendations or change your settings. That boundary is enforced at the tool level — an OAuth-connected agent that tries to accept a recommendation gets a structured denial with a redirect to the dashboard.

THE GOVERNED HANDSHAKE

TRY IT NOW

Paste this URL into any MCP-compatible client. The governance protocol takes care of the rest.