Watchful StatePowered by Insight Bridge
§ MCP · Model Context Protocol

Give an agent the whole record

Every number and quote on these pages comes from a live query, and the same record is open to AI agents through a Model Context Protocol server. Add it as a connector and your agent can search every passage, read every topic and every issuing body, compare what each kind of body says and follow every claim to its report.

Add it as a connector

Connector URLhttps://auditcorpus.aicolab.org/mcp

This report is public: no authentication is required.

Claude — personal accounts (Free, Pro and Max)
  1. Open Customize → Connectors.
  2. Click +, then Add custom connector; name it and paste the URL above.
  3. Leave Advanced settings empty and click Add.
  4. In a chat, click +, open Connectors and switch it on for that conversation.

Connectors are account-wide: added once, there in Claude on the web, desktop and mobile. Claude's connector guide →

Claude — organisation accounts (Team and Enterprise)

On Team and Enterprise plans only an organisation owner can add a custom connector; members do not see the option.

  1. An owner opens Organisation settings → Connectors and adds a custom web connector with the URL above.
  2. No authentication is required for this public report.
  3. Each member connects it from their own connector settings and switches it on in a chat.

Custom connector administration →

ChatGPT — custom MCP app
  1. Turn on Developer mode in Settings → Apps → Advanced settings, where your plan offers it; a workspace may need an administrator.
  2. Open Apps → Create and enter the URL above.
  3. Choose No authentication, scan the tools and create the app.
  4. Select the app in a conversation and ask about the record.

OpenAI's setup guide →

Microsoft 365 Copilot — via Copilot Studio
  1. Open an agent in Copilot Studio, then Tools → Add a tool → New tool → Model Context Protocol.
  2. Enter a name, a description of this record and the URL above. Choose None for authentication.
  3. Create the connection and add it to the agent. Test a question, then publish it.

Microsoft's connection guide →

Claude Code

Add the report as a remote HTTP server:

claude mcp add --transport http audit-corpus https://auditcorpus.aicolab.org/mcp

Run /mcp in Claude Code to check the connection. Claude Code's MCP guide →

Then ask: “Using The Anatomy of a Watchful State, which topics have the widest reach, where do issuing bodies differ, and what verbatim quotes support their positions?”

Transport: Streamable HTTP, stateless JSON-RPC over POST https://auditcorpus.aicolab.org/mcp. The endpoint answers each request with JSON and keeps no SSE stream open.

How to read what comes back

  • Counts describe this record. These are the issuing bodies in it, whose published reports were collected; "N issuing bodies say X" measures the record, never a wider population.
  • Positions are relative to each topic's own proposition. Assurance, Minor, Significant, Systemic measure a stance toward one framing, not agreement with the report or with each other.
  • One lens: Kind of body. Perspectives compare what each value of it says; the tools that take it list its values.
  • Use the ids the tools return. topic_id is the topicClusterId a topic tool returns; issuing body and report tools return the ids their detail tools take.
  • Two scores per report. Severity and systemic reach are graded on two 1–5 rubrics, so they rank reports against each other and never against an official standard; severity_breakdown pivots them by any slice.
  • Responses are recorded only where a report gives one. Many genres (royal commissions, Productivity Commission inquiries) structurally do not, and are excluded rather than read as evasion.

Try the tools

Choose a tool, adjust its inputs and call the live server: the exact request an agent makes. The tools, their inputs and their example values come from the server itself.

Loading the tools…