1. Overview
These MCP / AI Connector Terms (“MCP Terms”) apply when you connect any third-party AI client — including Anthropic’s Claude, OpenAI’s ChatGPT, Cursor, or any other MCP-compatible tool — to Taqql via the Taqql MCP (Model Context Protocol) server. These terms supplement the Terms of Service and are incorporated by reference.
2. How the MCP connection works
The Taqql MCP server implements the Model Context Protocol and exposes a set of read/write tools scoped to your authorized Taqql organization. Access is granted via OAuth 2.1 authorization code flow. You must explicitly authorize each third-party AI client and may revoke that authorization at any time from the Taqql web application.
When a connected AI client calls a Taqql MCP tool, it acts with the permissions of the Taqql user who authorized the connection. The AI client cannot exceed those permissions.
3. Your responsibilities
By connecting a third-party AI client, you:
- Accept responsibility for the AI client’s actions taken on your behalf. Taqql treats actions by an authorized AI client identically to direct actions by the authorizing user, and all such actions are recorded in the audit log.
- Agree to review your AI client’s data access and use policies independently. Taqql is not responsible for how third-party AI providers handle data passed to them through prompts or tool responses.
- Acknowledge data exposure: when you ask an AI client a question about your Taqql data, that data is sent to the third-party AI provider’s systems. You are responsible for ensuring this is compliant with your own policies and applicable law.
- Must not use connected AI clients to circumvent Taqql’s access controls, extract data beyond your authorized scope, or perform actions not permitted under the Acceptable Use Policy.
4. Taqql’s MCP data handling
- Taqql’s MCP server only returns data within the authorized user’s organizational scope.
- All MCP tool calls are logged in Taqql’s immutable audit log.
- Taqql does not send data to third-party AI providers directly; your configured AI client sends queries to Taqql and then independently sends context to its AI backend. Taqql does not control this flow.
5. Supported and authorized clients
Taqql currently supports and has tested MCP connectivity with:
- Anthropic Claude (claude.ai / Claude API)
- OpenAI ChatGPT (via ChatGPT plugins / Custom GPTs / Responses API)
- Cursor IDE
Other MCP-compatible clients may work but are not officially supported. Taqql reserves the right to restrict access from clients that violate these terms or our AUP.
6. AI outputs disclaimer
Responses generated by connected AI clients are produced by third-party AI systems and do not constitute professional advice (financial, legal, engineering, or otherwise). AI outputs may be inaccurate, incomplete, or out of date. You are responsible for verifying AI outputs against source data before making decisions.
7. Modifications
We may update these MCP Terms as the MCP protocol and AI ecosystem evolve. We will provide notice of material changes.