StructureClerk

Governing Anthropic Claude agents over MCP: authority and proof

Anthropic proposed the Model Context Protocol, now the agent integration standard. StructureClerk exposes it natively: wiring the authority layer into a Claude agent fits in a configuration file, without writing a line of code.

The Model Context Protocol was born at Anthropic and became, under Linux Foundation governance, the standard by which agents reach tools and data. Claude agents speak it natively — and StructureClerk exposes an MCP server over Streamable HTTP, the transport the specification standardized.

This is the shortest integration path in this wiki: one configuration entry, and the agent gains regulatory capabilities. No code to write, no connector to build, no data to move.

The integration scenario

  1. 1

    Generate your organization's profile

    The free assessment (start here) establishes your jurisdictions, sector and data categories. That profile is what makes decisions specific to your company rather than generic.

  2. 2

    Add the MCP server

    One configuration entry is enough. Streamable HTTP is the transport the specification standardized — no polling, no session to maintain.

    {
      "mcpServers": {
        "structureclerk": {
          "url": "https://structureclerk.ca/api/mcp-http",
          "transport": "streamable-http"
        }
      }
    }
  3. 3

    Let the agent read, keep the decision in your code

    The 5 exposed tools cover jurisdiction mapping, compliance checking, roadmap, algorithmic assessment and the authority decision. The first four are safe as open access; for the last one, call it yourself before the sensitive action.

  4. 4

    Wire the four verbs

    ALLOW executes. APPROVE requests human validation. DENY refuses and returns the reason — which the agent can explain to the user in its own words. ESCALATE holds and notifies. Keep the evidence identifier with the execution trace.

The exposed tools

ToolWhat it doesOpen access?
compliance_checkAnalyzes an organization's regulatory exposureYes — read
jurisdiction_mapMaps the laws applying to a jurisdictionYes — read
compliance_roadmapProduces a phased action planYes — read
algorithmic_compliance_assessmentAssesses a system's algorithmic complianceYes — read
authority_decideRenders the advisory decision with signed evidenceNo — call from your code

The evidence, and why it travels without you

Every decision leaves Ed25519-signed evidence, chained to the previous one. It verifies on our verification page — publicly, without an account. That is the difference between an internal log only you can read and evidence your customer, auditor or regulator checks themselves.

And because the decision is jurisdictional, it carries what a model cannot infer alone: the same export can be permitted toward one jurisdiction and denied toward another. The engine evaluates twice — without then with the jurisdictional rules — and the stricter outcome wins.

+How do I connect a compliance MCP server to Claude?

Through a configuration entry pointing at our MCP endpoint over Streamable HTTP. No code to write: the agent discovers the exposed tools and can call them. The full walkthrough is described in our Claude Desktop and Law 25 guide.

+Do I need an API key?

For general regulatory reads, no — they are reachable without an account. For decisions personalized against your organization's profile, yes: that profile is the difference between a generic answer and one that accounts for your jurisdictions and sector.

+Why not let the agent call the decision itself?

You can, for reversible low-stakes actions. But for an irreversible one — an export, a send, a payment — a guardrail the agent may choose not to call protects nothing. Put the call in the action's execution path rather than in the tool list.

+Does the engine receive the conversation content?

No. The decision is rendered on the action's metadata: type, data categories, jurisdictions, sector, autonomy level. Neither the conversation, nor documents, nor customer data are transmitted.

Start with the profile

The assessment is what personalizes the decisions your agents receive. It is free, and only needs your website address.

Other integrations