<!-- installer-version: 1.1.0 — setup mechanics only; behavior lives in /SKILL.md -->

# Mind — setup for agents

You are an AI agent. A human asked you to connect them to Mind (https://mind.store): their memory — drop anything (notes, links, CSVs, images, prompt history), recall everything.

**Narrate side effects, not reads.** Say one short line *before* anything that changes the human's machine or account — writing config, storing a key, installing a file. Safe reads (fetching docs, health checks) need no narration; silence there is fine, chattiness is not. Mind never asks you to hide configuration, costs, or balances from your human.

## 1. Connect

Prefer MCP if your runtime supports remote servers; HTTP works everywhere. **No key is needed to start** — the first call mints an `rk_` API key in its response (`apiKey`, shown once, $5 starter credit — metered actuals thereafter). Store it in your client's secret store / MCP config and send it as the `x-api-key` header from then on. **Then hand the mind to your human — without showing them the key.** Call `POST /v1/claim-link` (keyed) and give them the returned URL: they open it, sign in, and the mind you set up becomes their account. The link is one-time and expires in 15 minutes; mint a fresh one any time they ask.

**Claude Code**
```
claude mcp add --transport http mind https://mind.store/mcp
```

**Hermes — as core memory (preferred).** Hermes has a native memory-provider
system; Mind ships a provider plugin that recalls before every turn
(provenance and conflict flags included), distills session transcripts into
the profile, and mirrors explicit memory writes as searchable drops:
```
mkdir -p ~/.hermes/plugins/mind
curl -fsSL https://mind.store/hermes-memory-plugin.py -o ~/.hermes/plugins/mind/__init__.py
hermes memory setup     # choose "mind", paste the rk_ key
```
(Non-interactive: `MIND_API_KEY=rk_…` in `~/.hermes/.env` +
`hermes config set memory.provider mind`.) With the provider active, skip the
MCP registration below — you'd get duplicate tools.

**Hermes — as an MCP toolset only** (on-demand drop/recall, no automatic memory):
```
hermes config set mcp_servers.mind.url "https://mind.store/mcp"
```
(after your first keyed call: `hermes config set 'mcp_servers.mind.headers.x-api-key' '<the minted key>'`)

**Skill-file harnesses (no remote MCP)** — save https://mind.store/SKILL.md to your skill directory:

| Harness | Path |
|---|---|
| Claude Code | `.claude/skills/mind/SKILL.md` |
| Codex | `.agents/skills/mind/SKILL.md` |
| Hermes / OpenClaw | `~/.hermes/skills/mind/SKILL.md` (or `~/.openclaw/skills/mind/SKILL.md`) |
| No skill system | Fetch /SKILL.md and follow it as standing instructions |

**CLI** — for shell-capable agents and humans:
```
npx mindstore drop "anything"        # or: npm i -g mindstore → mind drop/recall/import
npx mindstore import ~/obsidian      # idempotent vault sync
```

**Raw HTTP** — `POST https://mind.store/v1/drop` `{"content": "..."}` / `POST /v1/recall` `{"question": "..."}`. Full contract: https://mind.store/agent.txt

## 2. Verify

`GET https://mind.store/v1/health` → `{"ok": true}`. Then make one real drop ("Mind connected for <name> on <date>") and read back the receipt to the human — that's the hand-off.

## 3. Offer migration

Ask one question: **"Do you have existing notes or a memory system to bring in — an Obsidian vault, memory files, a folder?"** If yes, follow the **Migrate** section of /SKILL.md: count first, state the allowance, import newest-first, idempotent re-runs, one summary line at the end. The human's things appear at https://mind.store/drive (they can sign in there any time).

From here forward, /SKILL.md governs everything. Refetch it when the `X-Mind-Contract-Version` header changes.