MCP Server for AI Governance: Let Any Agent Query AI Law

0:00One source of AI law for every agent

Welcome! Your company will build many AI agents. Should each one learn AI law from scratch? No. AI Governance Chain offers a free MCP server for AI governance. Any MCP-capable agent can ask which rules apply, what's due next, and whether your file passes.

Quick note: this is my AI twin. I write and approve everything it says, and every source is in the description. I am Mathieu K. Gouanou, a consultant based in Geneva, Switzerland, with 20+ years in organizational transformation. I write the AI Governance Chain newsletter on LinkedIn, and I am the founder of aigovernancechain.com. this is the Season 1 finale.

0:38The problem: AI law pasted into prompts

Here's a pattern I expect in many companies. One team builds a sales assistant. Another builds an HR helper. A third builds a support agent. Each one needs to know which AI rules apply. So each team copies some text into a prompt.

Six months later, the dates have moved. Remember Episode 4: the EU moved its own high-risk dates in July 2026. Nobody updates the prompts. Now you have three agents, with three different, outdated versions of the law. That's not a knowledge problem. It's an infrastructure problem.

1:08What is MCP?

Infrastructure needs a standard. The Model Context Protocol, MCP, is the open standard that connects AI models to tools and data. Anthropic launched it in November 2024. In December 2025, it was donated to the Agentic AI Foundation under the Linux Foundation. By then, more than 10,000 MCP servers had been published, and it was supported by Claude, ChatGPT, Microsoft Copilot, Gemini, Cursor, VS Code and others.

So here's the definition. The AI Governance Chain MCP server is a free, read-only connector that lets any MCP-capable agent query the Obligation Atlas and check AI_GOVERNANCE.md files, with a source for every answer.

1:51The 8 tools, step by step

It has 8 tools, all read-only. Step 1: scope_check. Give it the countries, the role and the use, and it returns the obligations that apply. Step 2: upcoming_deadlines. Ask what's due in the EU, and it returns 2 December 2026: the marking duty under Article 50(2) for generative systems already on the market, with its source.

Step 3: look things up. atlas_search finds obligations by keyword. Three more tools fetch one obligation, one regime, or the list of regimes. Step 4: work with files. check_ai_governance_md runs the 16 checks, and generate_ai_governance_md drafts a new file. On the chain, that covers LAW, OBLIGATION and the GATE.

To connect it, you add one short block to your MCP config in Claude, Cursor or VS Code. One command, link below.

2:50A real question, end to end

Here's what that looks like in practice. A platform engineer asks the company's internal agent: we're launching a chatbot with agent features in the EU and California. What applies? The agent calls scope_check. Back come 9 obligations, from the EU AI Act's Article 50(1) disclosure to California's BOT Act, each with a citation.

Then she asks: anything due soon in California? upcoming_deadlines answers with dates that include SB 1119's child-safety rules on 1 January 2027. Then: draft our file. generate_ai_governance_md writes a first AI_GOVERNANCE.md. Three questions, three cited answers, and nothing pasted into a prompt.

3:36Agent Skills or MCP: which one?

Episode 1 gave you Agent Skills. So why MCP as well? Skills put the knowledge inside agents that support skills. They even work offline. MCP lets any MCP-capable agent ask live, including the agents your company builds itself. Same Atlas underneath, two doors in. Use skills for coding agents. Use MCP for everything else.

3:57Who the MCP server is for

Agent builders, who want one cited source instead of pasted text. Platform teams, who want every internal agent to ask the same questions in the same way. Governance and risk tool vendors, who can build on open data instead of rebuilding it. And public bodies building assistants for staff, who need answers with sources they can check.

4:16Limits: read-only, draft data

The limits. The server reads. It doesn't change your files, and it runs on your machine over standard input and output. The answers are draft data, as of 4 November 2026, and every regime file stays draft-unreviewed until a named expert signs it off. Every answer carries a source, so check it. This is governance guidance, not legal advice in any jurisdiction.

4:38Your next step and Season 2

Your one action: add the server to your agent today, then ask it about your own product. The link is in the description.

That's Season 1: rules your agents can read. Season 2 goes under the hood. First up: what happens when an agent tries to refund 10,000 euros?

Thank you for watching!