# Quable MCP server [!badge variant="warning" text="PRIVATE BETA"]

!!!danger Private beta: access through the IAMP platform only
The Quable MCP server is a **private beta** and **not a public API**. Access is only authorized through the **IAMP platform**: any other access is not authorized. **Any tool, prompt, resource, parameter or behavior can change or be removed at any time, without notice.**
!!!

The **Quable MCP server** connects the AI agents of the **IAMP platform** to your Quable PIM & DAM.
It implements the [Model Context Protocol](https://modelcontextprotocol.io) (MCP), an open standard that lets an AI assistant discover and call *tools* exposed by a server.

From the IAMP platform, you can ask your assistant, in plain language, to search your catalog, read or enrich product data, tag assets, link media to products or run catalog exports. The assistant picks the right tools and calls the [Quable REST API](/quable-api/get-started/) on your behalf, with your API token.

---

## What you can do

||| :icon-search: **Explore your catalog**
Search products, variants, assets and classifications. Read attributes, completeness scores, links and workflows.

||| :icon-pencil: **Enrich your data**
Update product, variant and asset attributes. Create products and variants. Manage select options.

||| :icon-link: **Manage links**
Attach assets to products, find orphan assets, create cross-sell or upsell links.

||| :icon-download: **Export**
Run catalog exports and get the CSV or XLSX file back in the conversation.

|||

**Example requests:**

```text
Find all references with an e-commerce completeness below 80% in French, sorted from the least complete.
```

```text
List the assets uploaded this week that are not linked to any product, and suggest which product each one belongs to.
```

```text
Translate the English description of product REF-001 into German and Spanish, show me the result, then save it.
```

---

## Key facts

| | |
| --- | --- |
| **Status** | Private beta. Not a public API: tools and behavior can change at any time, without notice. |
| **Access** | Through the IAMP platform only |
| **Authentication** | Two HTTP headers: `X-Quable-Instance` and `Authorization: Bearer <token>` |
| **Token** | A Quable API token, created in **System → API token** (see [Authentication](/quable-api/authentication/)) |
| **Tools** | 35 tools at the time of writing: catalog structure, products, variants, assets, select options, links, exports |
| **Data source** | The Quable REST API v5 of your instance. The MCP server stores no catalog data. |

---

## How it works

```text
┌─────────────────┐  MCP (HTTP)   ┌──────────────────────┐  REST API v5  ┌─────────────────┐
│  IAMP platform  │ ────────────► │  Quable MCP server   │ ────────────► │  your-instance  │
│   (AI agents)   │ ◄──────────── │                      │ ◄──────────── │   .quable.com   │
└─────────────────┘ tool results  └──────────────────────┘               └─────────────────┘
```

1. The IAMP platform sends each request with your instance name and API token in HTTP headers.
2. The MCP server validates the token against your instance before running any tool.
3. Each tool translates the call into one or more Quable REST API requests and returns a result shaped for an AI model: compact, with exact codes, and with explicit errors instead of silent failures.

The server is **stateless**: no session is kept between requests and no catalog data is stored.

---

## Security and permissions

- **The token decides what the assistant can do.** A *Read access* token only allows reading: every write tool fails. Use a *Full access* token only when you want the assistant to modify data.
- **Write tools ask for confirmation.** Every tool that creates, updates or deletes data tells the assistant to confirm with you before calling it. Read the proposed change before you approve it.
- **Writes are checked.** Write tools refuse values the PIM would drop or garble, and read back what was stored, so the assistant reports what was actually saved.
- **Your data stays out of the logs.** The MCP server never logs tool results, and only logs argument *names* and sizes, not their values, except for identifiers such as product codes and locales.

!!!warning
An AI assistant can make mistakes. Test on a sandbox or staging instance first, and keep write access for the people and use cases that need it.
!!!

---

## Next steps

- [:icon-plug: Access through IAMP](/mcp-server/connect/)
- [:icon-tools: Tools reference](/mcp-server/tools/)
- [:icon-workflow: Workflows and example prompts](/mcp-server/workflows/)
- [:icon-alert: Limits and troubleshooting](/mcp-server/limits-and-troubleshooting/)
