Quickstart

Three steps take you from nothing to a coding agent that can buy data on the Amnetic marketplace: connect an MCP client, then drive the enter_marketpurchase loop. Everything runs over MCP — there is no CLI to install.

1. Connect your MCP client

Sign up on the website (the marketplace is invite-gated), mint an amn_ API key in the portal, and point your client at the endpoint. With Claude Code that's one command:

claude mcp add --transport http amnetic \
  https://market.amnetic.ai/mcp \
  --header "Authorization: Bearer amn_YOURKEY"

The full walkthrough — including the claude.ai browser connector, which uses OAuth and needs no key — is on Connect your MCP client.

2. Add credit

New accounts start at a $0 balance, and the same wallet funds both running a market-entry session and wallet purchases. Top up credit in the portal (Stripe Checkout), or have the exchange grant credit during onboarding / design-partner setup. For a specific listing, the portal and REST API can also use card checkout at purchase time while leaving wallet credit available for operations. Check your credit any time with the read-only balance MCP tool — there is no MCP top-up tool, so funding happens in the portal.

3. Use it from Claude Code

Your agent now has the 13 core MCP tools — the buyer loop is enter_marketpurchase, with balance, list_images, ownership_list, ownership_download, and purchase_status around it (see the MCP tools reference for all of them). Ask it to find and buy data, and it will hand the goal into the walled exchange, get a recommendation, and buy exactly what you approve:

Find me a dataset of US retail foot-traffic, weekly granularity, under $50, then buy the best match.

Under the hood that calls enter_market (a forgetful inner agent evaluates real seller data inside the wall and returns a buy recommendation — free and read-only), you decide, purchase buys the listings, and ownership_download retrieves the owned bytes. For the tool shapes and per-client claude mcp add paths, see Connect your agent.

Once a deployment enables license terms, every purchase also records a license grant you can read back: ownership_list carries a per-listing license block, and amnetic buyer license <listing-id> prints the exact operative license text-of-record (add --json for the pinned hashes). Before purchase, an open listing's offer_key is readable at GET /api/v1/listings/{listing-id}/license-text/{offer-key} with the reviewed terms, hash, and settlement token. These read surfaces are dark until the licensing terms are counsel-ratified; restricted-listing availability and legal acceptance sufficiency remain pending counsel review.

Credits

Your account has one balance that funds everything an enter_market run touches: the pre-run affordability check, the inner-agent session (compute + LLM run-cost), and the final purchase debit. Fund it with a portal Stripe top-up, or have the exchange grant credit during onboarding — either way the same wallet pays for both running and buying.

balance        →  { "balance_cents": 4200, "currency": "USD" }   # read-only MCP tool

The balance tool is read-only: it reports your spendable credit but cannot add to it. If balance returns 0, fund the account in the portal or ask the exchange to grant onboarding credit before running the loop. See Connect your agent for the full setup flow.

Going deeper (advanced / SDK integrators)

The buyer path above (enter_marketpurchaseownership_download) is the whole supported flow. Integrators who can't use MCP — or who need raw session controls, a custom buyer-image push, or the signed audit record — drive the equivalent REST + SSE surface directly. That surface is not the buyer path; it's the advanced fallback, documented in the API reference.

See the MCP tools reference for every tool and its parameters, the API reference for the advanced REST + SSE shapes, and How it works for the concepts.