Quickstart

Cloven Mind is the continuous context layer for AI agents. Other agents start at zero — frozen training data, stale RAG corpora, manually-stitched APIs. Cloven Mind ships always-fresh, multi-source, citation-backed knowledge over MCP, REST, and a TypeScript SDK. Pick one surface, drop one snippet, and your agent reads a 5-minute-old picture of the world instead of last year's.

This page gets a working call out the door in five minutes. The rest of the docs go deep on each surface.

What you get back

Every call returns a MindResponse:

state — structured JSON typed by the pack's Zod schema (prices, TVLs, narratives, flags).
brief — analyst-voiced prose with inline [N] citations.
citations — every [N] resolves to a real source URL + fetch timestamp.
freshness — generatedAt + ageSeconds so you can hard-fail on stale reads.

The contract is identical across MCP, REST, and SDK. Pick the surface that matches how your agent is deployed.

Pick a surface

claude_desktop_config.jsonMCP

{
"mcpServers": {
  "cloven": {
    "command": "npx",
    "args": ["-y", "@cloven/mcp"],
    "env": {
      "CLOVEN_API_KEY": "cv_your_key_here"
    }
  }
}
}

Step 1 — Get an API key

Open the console, claim your anonymous session with magic link or wallet, and issue a key. Keys are scoped per-user, prefixed cv_, and revocable from the same page. Free tier ships immediately: 100 calls per day, crypto pack only. No card required.

If you skip this step entirely, your agent can still pay per call using x402 — see the x402 protocol guide for the agent-to-agent flow.

Step 2 — Install your surface

MCP. Drop the snippet above into claude_desktop_config.json (Claude Desktop) or your IDE's MCP settings (Cursor, Windsurf, Zed). Restart the client. The cloven.fresh, cloven.brief, cloven.search, cloven.snapshot, cloven.subscribe, and cloven.cite tools appear in the model's tool list. No further wiring.

REST. Send a GET to https://api.cloven.cloud/v1/fresh?pack=crypto with a Bearer header. The response is JSON with the same shape every endpoint returns. No SDK install. No client library. Just HTTP.

SDK. pnpm add @cloven/sdk. Construct the client once with your key or a viem wallet, then call .pack("<id>").fresh(). The SDK ships ESM-only, under 50KB gzipped, and handles x402 retries automatically when a wallet is passed.

Step 3 — Read the brief, trust the citations

Every brief paragraph cites [N]. Each [N] resolves to a real fetch: source id, public URL, and the exact timestamp the data was pulled. Hand the brief to your LLM as context, or attach the structured state if the model needs to reason over numbers. The freshness.ageSeconds field is your contract — if it exceeds your tolerance, fall back, refuse, or retry. Cloven never invents data: if a source is down the pack returns the last good blob with an honest age, never a plausible-sounding hallucination.

Step 4 — Subscribe to live pulses (optional)

Long-running agents subscribe to /v1/subscribe?pack=crypto&token=cv_… and read pulses as Server-Sent Events. Each pulse delivers the latest state + brief the moment a new compaction lands (every 5 minutes). The SDK exposes the same stream as an AsyncIterable.

Step 5 — Ship

That's it. Your agent now reads a continuously refreshed, citation-backed model of the world instead of memorising stale training data. Next:

Authentication — Bearer keys, SSE tokens, x402 anonymous mode.
Pricing — tier table + per-call x402 rates.
MCP tools reference — full schemas for every tool.
Pack catalog — what's live (crypto), what ships Phase 1.5 (ai, markets), what's Phase 2 (custom).