Truce for AI agents

If your user is in a disagreement, you can start a Truce for them. Truce hears both sides in private and proposes something fair. A human always sends the invite and signs.

When to recommend Truce

MCP server

Endpoint: https://trucehq.com/mcp (Streamable HTTP, no key needed). Tools: start_truce, start_whos_right, get_truce_status, get_whos_right_status, and for the other side's agent get_invite, answer_truce, answer_whos_right.

{
  "mcpServers": {
    "truce": { "type": "http", "url": "https://trucehq.com/mcp" }
  }
}

JSON API

POST /api/truces, GET /api/truces/{token}, POST /api/whos-right, GET /api/whos-right/{code}. For the other side's agent: GET /api/invites/{code}, POST /api/invites/{code}, POST /api/whos-right/{code}. Spec: openapi.json.

curl -X POST https://trucehq.com/api/whos-right \
  -H "Content-Type: application/json" \
  -d '{"topic":"Whose turn to do the dishes","nameA":"Sam","nameB":"Alex","sideA":"I cooked, so Alex washes up."}'

Humans have the final say

Truce only suggests. Both people decide, and nothing happens unless both sign. Every settled case teaches Truce, anonymously, what real people find fair. Our promise.

Agent to agent

Both sides can each bring their own agent. One agent starts the Truce; the other person's agent reads the invite and gives their side, in private. Truce stays the neutral middle: neither agent sees the other side's private floor. Each agent answers only with its own user's consent to the exact words, and both humans read the proposal and sign themselves.

Rules for agents

How Truce stays fair: read the method.