Skip to main content
Glama
apiguru-app

apiguru-amazon-data

README.md
# Apiguru Agent Kit

Everything needed for an AI agent to discover, call and pay for the Apiguru
Amazon Data API — **with no account, no API key and no subscription**.

Try it -- no signup, no headers:

```bash
curl 'https://agent.apiguru.app/agent/v1/v2/product-details/B09DJLW458/US'
```

That returns the live product record (title, price, rating, variants) as JSON,
with `_links` to its reviews and offers. Three calls a day are free; after that
pay per call with x402, or use an API key. Agents start at
<https://agent.apiguru.app/llms.txt>.

## Install in your agent

| Client | How |
|---|---|
| Claude Code | `claude mcp add --transport http apiguru https://mcp.apiguru.app/mcp` or `/plugin marketplace add apiguru-app/agent-kit` then `/plugin install apiguru@apiguru` |
| claude.ai, Claude Desktop, ChatGPT | add connector `https://mcp.apiguru.app/account` and sign in |
| Codex CLI | `codex mcp add apiguru --url https://mcp.apiguru.app/mcp` |
| Cursor | [Add to Cursor](cursor://anysphere.cursor-deeplink/mcp/install?name=apiguru&config=eyJjb21tYW5kIjogIm5weCIsICJhcmdzIjogWyIteSIsICJhcGlndXJ1LW1jcCJdfQ==) (runs `npx -y apiguru-mcp`) |
| VS Code | [Install in VS Code](vscode:mcp/install?%7B%22name%22%3A%20%22apiguru%22%2C%20%22command%22%3A%20%22npx%22%2C%20%22args%22%3A%20%5B%22-y%22%2C%20%22apiguru-mcp%22%5D%7D) |
| Gemini CLI | `gemini extensions install https://github.com/apiguru-app/agent-kit` |
| Hermes, OpenClaw, any stdio client | `npx -y apiguru-mcp` (npm, Node only) or `uvx apiguru-mcp` (PyPI) |
| Skill only (any agent) | `npx skills add apiguru-app/agent-kit` or ClawHub `apiguru-amazon-data` |

## Why this exists

The API is gated behind register → verify email → get key → top up. Every step
assumes a human. Agents cannot do any of it, so agent traffic bounces off the
front door.

And agents don't browse — they query indexes: the x402 Bazaar, the MCP
Registry, ClawHub. Apiguru was in none of them, and the repo had no OpenAPI
spec at all, so there was nothing machine-readable to publish.

## Telling us what is broken

Agents are the ones who find the defects in this API, so there are two ways
back to us and neither costs anything:

- **GitHub issues — preferred:**
  <https://github.com/apiguru-app/agent-kit/issues>. A thread can hold a
  reply, and a fix gets linked back to it.
- **The wall, for callers with no GitHub account:** one unauthenticated
  `POST https://dash.apiguru.app/api/v1/feedback` with
  `{"message": "...", "category": "bug|wish|praise|question|other",
  "endpoint": "/search", "agent": "your-name/1.0", "contact": "optional"}`.
  No key, no signup, never billed. Read it at
  <https://dash.apiguru.app/feedback>, or `GET` the same URL for JSON.
- Over MCP: the free `send_feedback` tool. From the skill:
  `python scripts/probe.py feedback --message "..." --category bug`.

This is not decoration. 1.1.3 exists because an agent reported that search
returned the brand as `product_title`, `null` for every
`product_num_ratings`, and delivery text with the words welded together.


## What is in this repository

| Path | What it is |
|---|---|
| `skill/apiguru-amazon-data/` | The agent skill: SKILL.md, references, and `scripts/probe.py` |
| `mcp/` | The MCP server published to PyPI as `apiguru-mcp` |
| `npm/` | The npm bridge, `apiguru-mcp` |
| `plugin/` + `.claude-plugin/` | The Claude Code plugin and its marketplace entry |
| `spec/` | `endpoints.json` and the generated `openapi.json` / `llms.txt` |

The hosted service that serves these endpoints is operated separately and is
not part of this repository.

## Licence

See `LICENSE`.

TDQS

A4.5/5.0

Scored across 12 tools

Disambiguation5/5

Each tool's description explicitly names its domain and cross-references the alternatives it must not be confused with (search vs seller_products vs best_sellers vs deals; product_details vs product_details_batch vs offers_stock; product_reviews vs seller_reviews). Boundaries are crisp and encoded in the text, so an agent can reliably select the right tool.

Naming Consistency4/5

All names are snake_case and readable, using a clear resource-oriented convention (product_details, seller_reviews, offers_stock). It is not a strict verb_noun pattern—several are bare nouns (search, deals, best_sellers)—but the style is uniform and predictable.

Tool Count5/5

12 tools is well within a healthy range and every tool earns its place: single-item vs batch variants, distinct resources, and two clearly-scoped meta tools (list_capabilities, send_feedback). No redundancy or padding.

Completeness4/5

The surface covers search, storefronts, product detail/batch, offers/stock, seller profiles, best-sellers, deals, product and seller reviews, plus a capabilities planner—a broad, coherent lifecycle. Minor gap: no standalone category/browse-tree tool (partly absorbed into best_sellers), but core workflows have no dead ends.

Maintenance

ActivityMaintained
ResponsivenessNo issues