# Periskop – product discovery (MCP/API) for AI agents

> Source: <https://www.periskop.ai>
> Published: 2026-06-30 20:40:07+00:00

# Product discovery for AI agents.

Cheap, reliable, hallucination-free product discovery for the agents you're building. One MCP/API call turns shopping intent into real, in-stock products, with live prices and working merchant links.

Example output · hover to pause, or pick an answer_type

## Use Periskop through MCP or REST.

Connect the MCP server to Claude Code, Cursor, and other ChatGPT-compatible or custom agent runtimes, or call the REST endpoint from any app backend. Keys, usage, billing, docs, and OpenAPI live in the developer portal.

```
// MCP, Claude Code, Cursor, custom agent runtimes
{
  "mcpServers": {
    "periskop": {
      "url": "https://mcp.periskop.ai",
      "headers": { "Authorization": "Bearer dp_..." }
    }
  }
}
// REST, from any app backend
POST /shopping/discover
Authorization: Bearer dp_...
{ "prompt": "...", "mode": "bundle" }
```

Create a project, mint a key, and call run_shopping_discovery. Every result carries a result_id, re-fetch it with get_discovery_result.

Drop Periskop into your existing automation stack, native steps for Dify, n8n, and Pipedream.

## Search returns links. Periskop returns the right products.

A generic agent searches, opens page after page, and re-reads it with an LLM on every request, slow, token-heavy, unreliable. Periskop returns the products in one structured call, with a paste-ready reply your agent renders directly.

## Onboarded does not mean trusted.

Periskop validates store structure, product extraction, price readability, link validity, image presence, and stability before a store can influence recommendations. Stores earn the right to shape recommendations.

- Accessible
- Product pages found
- Product cards detected

- Product extraction
- Price readability
- Link validity
- Image presence
- Stability / latency

**Trusted**, can shape best picks** Limited**, alternatives or restricted contexts** Rejected**, cannot influence recommendations

## Checkout was always solved. Knowing what to buy wasn't, until Periskop.

Payments, carts, and fulfillment already have a hundred providers, that part was never the problem. The unanswered question is the valuable one: out of everything for sale, what should your agent actually recommend? Periskop is that layer. It finds the right in-stock products and hands off a clean merchant link; your existing rails do the rest.

- Intent understanding
- Retrieval
- Ranking
- Alternatives
- Bundles
- Caveats
- Match quality
- Merchant links

Stripe · merchants · your storefront

- Checkout
- Payments
- Orders
- Fulfillment
- Returns

## One call. Build whatever needs live products and prices.

The same run_shopping_discovery call powers a chat assistant, a computer-use agent, a price-monitoring bot, an arbitrage scanner, a procurement workflow, a commerce copilot, a whole shopping app, and whatever you imagine next. You ship the experience; Periskop is the one call underneath.

Real, in-stock picks dropped straight into chat, ranked, priced, and linked.

[See pricing →](/pricing)

## Product discovery for the agents you are building.

For AI shopping assistants, browser and computer-use agents, procurement workflows, commerce copilots, recommendation systems, and AI-native shopping apps.

## Common questions

## What is Periskop?+

Periskop is product discovery for AI agents. One MCP/API call turns a shopping request into real, in-stock products, with live prices, working merchant links, alternatives, inferred bundles, caveats, a paste-ready reply, and a result_id, so your agent can recommend instead of crawling the web on every request.

## Who is this for?+

Developers building AI shopping assistants, browser and computer-use agents, procurement and ops workflows, commerce copilots, recommendation features, accessibility shopping, and AI-native shopping apps.

## How does it work?+

You send a prompt to run_shopping_discovery over MCP or REST. Periskop retrieves and ranks across quality-gated merchants and returns a structured response, items or an inferred bundle, plus a paste-ready suggested_agent_reply and next_actions, your agent renders directly. No per-site browsing, no runtime scraping.

## Do merchants need to integrate?+

No. Periskop validates store structure, extraction, prices, links, images, and stability so only quality-gated, discovery-ready supply can influence results. Merchants don't lift a finger.

## Is this just search?+

No. Search returns links you still have to open and reason over. Periskop returns the products themselves, ranked, in-stock, priced, and linked, in one call, with match_quality and caveats.

## Do you handle checkout or carts?+

No. No checkout, no payments, merchant links only. Periskop never buys, places orders, or creates merchant checkout carts. Users complete purchases on merchant sites.

## What happens when no good product exists?+

Periskop marks items match_quality: null and returns caveats rather than a forced or hallucinated pick, plus the closest matches and alternatives. Every result carries a result_id, so you can re-fetch it later with get_discovery_result.

## What is the MCP/API?+

An MCP server for agent runtimes (Claude Code, Cursor, custom agents) and a REST endpoint for app backends. Keys, usage, billing, docs, and OpenAPI live in the developer portal.

## How does pricing work?+

Usage-based: €7 per 1,000 shopping requests, with free credits on your first API key. Early access is founder-led onboarding.
