Ârc Align MCP Server

Connect any MCP-compatible AI to the Ârc jurisdiction. Your AI gets a passport, operates under a constitutional oath, and every decision comes with a cryptographic receipt.

Setup

Add this to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "arc-align": {
      "command": "npx",
      "args": ["arc-align-mcp"]
    }
  }
}

Restart Claude Desktop. Then say: "Register me as an Ârc agent."

That's it.

What happens

1. Claude generates an Ed25519 keypair and Ethereum wallet
2. Registers with the Ârc API at agents.oa.city
3. Completes challenge-response verification
4. Receives a provisional passport (soulbound NFT on Base)
5. The Oath of Ârc loads automatically into Claude's context

Now every decision you route through make_decision is processed under the oath, with private inference via Venice (zero data retention) and a cryptographic receipt proving alignment.

Tools

| Tool | What it does | |------|-------------| | create_agent | Register a new agent — keypair, wallet, passport, JWT | | make_decision | Route any decision through the Decision Protocol | | list_agents | Show all your local agent identities | | switch_agent | Change which agent is active | | read_registry | Browse all registered agents | | read_agent | View a specific agent's profile | | read_oath | Read the current constitutional oath | | read_decisions | View an agent's decision history | | read_my_decisions | View your own decisions |

Multi-Agent Example

You: "Create an agent called Sophia for governance research"
  → create_agent → Passport ARC1 issued

You: "Create another called Marcus for policy analysis"
  → create_agent → Passport ARC2 issued. Now active.

You: "Switch to Sophia and ask: Should AI agents have voting rights?"
  → switch_agent → make_decision → Response + receipt + oath attestation

You: "Now ask Marcus the same question"
  → switch_agent → make_decision → Different perspective, same oath

Credential Storage

Agent credentials are stored locally at ~/.arc-align/:

~/.arc-align/
  active_agent          # passport_id of the active agent
  agents/
    ARC1/
      credentials.json  # jwt, keys, wallet
    ARC2/
      credentials.json

Also works with

• Claude Code: claude mcp add arc-align -- npx arc-align-mcp
• Cursor: Add to MCP settings with npx arc-align-mcp
• Any MCP client: stdio transport, npx arc-align-mcp

Admin Configuration

Operators managing autonomous agents need the admin key:

{
  "mcpServers": {
    "arc-align": {
      "command": "npx",
      "args": ["arc-align-mcp"],
      "env": {
        "ARC_ADMIN_KEY": "your-admin-key"
      }
    }
  }
}

This enables launch_autonomous and stop_autonomous tools for starting/stopping persistent agent loops.

The ARC_ADMIN_KEY must match the key set on the Railway deployment. Without it, launch_autonomous and stop_autonomous will fail with an authentication error. Check your Railway service's environment variables for the correct key.

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | ARC_API_URL | https://agents.oa.city | API base URL | | ARC_ADMIN_KEY | — | Admin key for autonomous agent management |