# BondPrint — Guide for AI Agents

BondPrint (https://bondprint.ai) is a privacy-first relationship-intelligence app. It helps a person build a living map of how they love — through a baseline assessment, adaptive check-ins, tentative hypotheses with evidence, a reflection guide, a journal, and an optional couple comparison. Tagline: "Love isn't random. It's patterned."

This document explains how an AI agent can act on a BondPrint user's behalf, with their explicit consent.

## Voice and stance

BondPrint speaks carefully and tentatively. Readings are tendencies, never diagnoses. Hypotheses are working guesses the user can reject. If you relay BondPrint content to the user, keep that stance: prefer "you tend to", "this might", "worth noticing" over clinical or absolute claims. BondPrint is not therapy and never labels people.

## MCP server

BondPrint exposes a Model Context Protocol (MCP) server over Streamable HTTP:

- Endpoint: `https://bondprint.ai/api/mcp`
- Transport: Streamable HTTP (JSON-RPC over POST; sessionless)
- Auth: `Authorization: Bearer <token>` — a personal access token the user creates in **Settings → Connect an AI assistant**

The token grants the same access as the user's own session for the tools below. Users can revoke it at any time in Settings.

### Example client configuration

```json
{
  "mcpServers": {
    "bondprint": {
      "type": "http",
      "url": "https://bondprint.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer bp_agent_..."
      }
    }
  }
}
```

### Available tools

Read:
- `get_profile` — completeness, current-state narrative, and all domains with dimension readings (band, score 0-100, confidence)
- `get_next_questions` — adaptively selected check-in questions (may be empty; respects the user's daily limit and pause setting)
- `list_hypotheses` — tentative hypotheses with status and confidence
- `list_tensions` — productive tensions (places the user's readings pull in different directions)
- `list_journal_entries` — the user's journal, newest first
- `get_couple_status` — whether the user has linked BondPrints with a partner
- `get_settings` — pacing and guide-voice settings

Write:
- `submit_answer` — record an answer to a check-in question. Only submit answers the user actually gave or clearly endorsed; never invent answers.
- `add_journal_entry` — save journal text the user wrote or asked to record (supports `noAnalysis` to keep it out of profiling)
- `update_settings` — pacing and guide-voice preferences only

### Deliberately not available to agents

- Deleting data or the account
- Signing in/out or changing the account email
- Creating, accepting, or breaking a couple link
- The user's raw reflection (AI guide) conversations

These stay in the app, under the user's own hands.

## Privacy boundaries

- All data belongs to the user; a token only ever reads/writes that one user's data.
- Journal entries are private. Quote them back sparingly and only in service of what the user asked.
- Couple comparisons never expose journal entries or raw answers — only dimension readings.
- If the user shares something suggesting harm to themselves or others, encourage them toward real human support; BondPrint is not a crisis service.

## Without MCP

If you cannot use MCP, ask the user to export their data (Settings → Export Data) and share the JSON with you directly.