# MCP FAQ

Source: https://userguide.qualytics.io/using-the-platform/mcp/faq/

Common questions about the Qualytics MCP server: what it is, which clients can connect, how access is authorized, and what a connected client can do. For something that is not working rather than not clear, see [MCP Troubleshooting](https://userguide.qualytics.io/using-the-platform/mcp/troubleshooting/).

## General

#### What is the Qualytics MCP server?

A remote endpoint that AI clients such as ChatGPT, Claude, and Cursor connect to through the [Model Context Protocol](https://modelcontextprotocol.io/). Once connected, the client can search your assets, read quality results, and take supported actions through the same tools AgentQ uses. See [How MCP Works](https://userguide.qualytics.io/using-the-platform/mcp/deep-dive/how-it-works/).

#### What is the difference between AgentQ, the MCP server, and agentic endpoints?

| Component | Purpose |
| --- | --- |
| **AgentQ** | The built-in Qualytics assistant, with its own chat and user interface. |
| **MCP server** | A remote protocol endpoint that external AI clients connect to through OAuth. |
| **Agentic endpoints** | REST endpoints for custom applications and automation. |

AgentQ and MCP clients use the same tool catalog. They manage conversations and authorization separately.

#### Do external clients have the same capabilities as AgentQ?

They discover the same Qualytics tool catalog, subject to the client's granted access and your Qualytics permissions. Each client has its own model, conversation history, interface, and way of asking you to approve tool calls. AgentQ's page context, suggestions, and PDF export belong to its built-in chat.

#### What tools are available?

The available tools depend on your Qualytics permissions and on the modules your organization has enabled. See the [Tool Catalog](https://userguide.qualytics.io/agentq/deep-dive/tool-catalog/) for the full list and the category of data each tool shares.

#### Does the server offer ready-made prompts?

Yes. The server ships with prompts such as **Investigate Anomaly**, **Generate Quality Check**, **Assess Data Trust**, and **Analyze Trends**. Your client lists them in its prompt or command menu, asks for a few inputs, and then follows the steps the prompt describes. See [MCP Examples](https://userguide.qualytics.io/using-the-platform/mcp/deep-dive/examples/).

---

## Connecting Clients

#### Which external AI clients are supported?

Any client that supports MCP over streamable HTTP and completes the authorization in a browser can connect. The guide includes setup steps for these clients:

| Client | Transport |
| --- | --- |
| Amazon [**Amazon Q Developer**](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/amazon/) (IDE) | Streamable HTTP |
| OpenAI [**ChatGPT**](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/openai/) | Streamable HTTP |
| Anthropic [**Claude Code**](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/anthropic/) | Streamable HTTP |
| Anthropic [**Claude Desktop**](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/anthropic/) | Streamable HTTP |
| OpenAI [**Codex**](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/openai/) (app, CLI, and IDE extension) | Streamable HTTP |
| Cursor [**Cursor**](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/cursor/) (editor and CLI) | Streamable HTTP |
| Google [**Gemini**](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/google/) (app and CLI) | Streamable HTTP |
| xAI [**Grok**](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/xai/) | Streamable HTTP |
| Amazon [**Kiro**](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/amazon/) (IDE and CLI) | Streamable HTTP |
| Microsoft [**VS Code (GitHub Copilot)**](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/microsoft/) | Streamable HTTP |
| Windsurf [**Windsurf**](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/windsurf/) (editor and plugins) and Devin CLI | Streamable HTTP |

For any other client, see [Other MCP-Compatible Clients](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/other-clients/).

#### What do I need before I connect a client?

A Qualytics Personal API Token and your Qualytics MCP URL, which is the address of your instance followed by `/api/mcp/`. See [External AI Client Requirements](https://userguide.qualytics.io/using-the-platform/mcp/connecting-external-ai-clients/requirements/).

#### Do I need to configure an AI provider in Qualytics to use an external client?

No. Each external client brings its own model. An AI provider is only required for AgentQ's built-in chat.

#### Why can't ChatGPT, Claude Desktop, Gemini, or Grok reach my Qualytics instance?

Those clients connect to MCP servers from their vendor's cloud, not from your computer, so your Qualytics instance must be reachable from the internet. Clients that run in your editor or terminal, such as Claude Code, Cursor, VS Code, Codex, and Kiro, connect from your own machine and work on a private network.

#### Do I have to authorize each client separately?

Yes. Each client keeps its own authorization, even when two clients read the same configuration file. Sign in once in each client you use.

---

## Access and Security

#### How is MCP access authorized?

Your client opens a Qualytics authorization page where you paste a Personal API Token. Qualytics then grants the client a separate MCP access token, and the API token itself never reaches the client. Every tool call is checked against the identity and permissions behind your token. See [MCP Permissions](https://userguide.qualytics.io/using-the-platform/mcp/deep-dive/permissions/).

#### Can I give a client read-only access?

Yes, when the client lets you request it. A read-only client can run the tools that read data but not the tools that change it. A client that requests no specific access receives read and write access, and the authorization page shows what was requested before you paste your token.

#### How long does an authorization last?

An authorization lasts 30 days, even when the client refreshes its connection. After that, the client asks you to connect again. It ends sooner if the Personal API Token behind it is revoked or expires. See the [MCP API](https://userguide.qualytics.io/using-the-platform/mcp/api/) for every token lifetime.

#### Can the MCP server write to my source datastores?

No. Tools can change Qualytics assets and can start operations that write to an [enrichment datastore](https://userguide.qualytics.io/source-datastore/enrichment-datastore/deep-dive/introduction/). They do not write back to source datastores. Review the actions your client proposes before approving them.

**Still have questions?**
Contact the Qualytics support team at
[support@qualytics.ai](mailto:support@qualytics.ai).

*Last updated: November 7, 2026.*