AI Connector

Time to complete: 5 minutes
Prerequisites: A Conveyor account.

Overview

The Conveyor AI Connector puts your customer trust program inside the AI tools your team already uses. Connect Claude, ChatGPT, Notion AI, Atlassian Rovo, Glean, or any other MCP-compatible assistant directly to your Conveyor account, and your team can answer security questions from knowledge you've approved, pull analytics in plain language, and keep your Knowledge Base current — all through natural conversation.

Under the hood, the AI Connector is Conveyor's hosted MCP server. MCP is an open standard that gives AI models a structured way to call external tools — similar to an API, but for AI assistants. Connect once and it works with any assistant that supports MCP: no custom integration, no API scripting, and setup takes seconds.

📘
What can I do with it?
Here are a few examples of things you can ask your AI assistant once connected:

"What questionnaires are due this week?"

"Approve all pending access requests from @acme.com"

"Do you encrypt data at rest?"

"Which customers downloaded our SOC 2 report in the last 30 days?"

See Using the AI Connector for full workflows.

Available Tools

Once connected, your AI assistant automatically discovers all available tools. These tools map directly to Conveyor's public API — the AI Connector provides the same capabilities, just accessible through natural language instead of raw API calls.

Because the AI Connector speaks standard MCP, it isn't limited to chat assistants — any system that can run an MCP client can connect, including AI platforms like Notion AI and Atlassian Rovo.

What your assistant can see and do is controlled by how you authenticate: with OAuth, your existing Conveyor permissions carry through, so the assistant only ever sees what you're cleared to see or edit. API tokens carry the fine-grained permissions you set when creating the token. In both cases, the AI Connector enforces the same roles and permissions as the Conveyor app — if you can't edit something in Conveyor, your assistant can't either.

Authentication

Mode	How it works	Best for
OAuth 2.1 (preferred)	Click Connect in your AI client, log in to Conveyor in the browser, and pick which organization to authorize. Permissions match your Conveyor user.	Claude, ChatGPT, any modern MCP client
API Token (Bearer)	Generate an API token in Conveyor — see Generating an API Token for instructions. Paste the token into your client's MCP config (e.g. `mcp.json`). Permissions are tied to the token's grants, which you can scope to specific resources and actions.	Cursor, OpenAI Playground, MCP Inspector

📘
What if my client supports both?
Prefer OAuth. It's tied to your Conveyor user — the assistant gets exactly the per-organization permissions you have in Conveyor. If a user's Conveyor access is removed, their connection stops working: the token can no longer refresh, and re-authentication is blocked.

Ready to connect? Follow Connecting Your AI Client for step-by-step setup in Claude, ChatGPT, and other clients.

Common Questions

Q: Do I need to self-host anything?
A: No. Conveyor hosts the AI Connector at mcp.conveyor.com/mcp. Just point your client to it and authenticate via OAuth or API token.

Q: Should I use OAuth or an API token?
A: OAuth if your client supports it — it's tied to your Conveyor user, so the assistant gets exactly the permissions you have and there's no long-lived bearer token sitting in a config file. Use API tokens for everything else, and scope them to only the resources and actions your workflow needs.

Q: Can the AI assistant access data outside my account?
A: No. The AI Connector uses the same roles and permissions as the Conveyor app, for both OAuth and API tokens. With OAuth, the assistant inherits the per-organization access of the logged-in user — if you don't have edit permissions in Conveyor, your assistant can't edit either. With an API token, the assistant can only access data the token is authorized to see.

Q: How do I limit or revoke access later?
A: Access follows the same permissions and roles you already manage in Conveyor — if someone's access changes there, it changes for their assistant too. Removing a user from Conveyor cuts off their connection entirely: the OAuth token can no longer refresh, and they can't re-authenticate. To disconnect a client yourself, remove the Conveyor connector in your AI client. For API tokens, revoke or re-scope the token in Conveyor at any time — see Generating an API Token.

Q: Can I see what's being asked and answered?
A: No. Activity through the AI Connector is captured in access logs, but not the specific questions and answers. The only activity that is tracked is user sessions and the tool calls that are used which Conveyor uses for performance monitoring.

Q: Is the AI Connector the same thing as Conveyor's MCP server?
A: Yes. AI Connector is the product name; MCP is the open standard it's built on. If your client asks for an "MCP server URL," use https://mcp.conveyor.com/mcp.

Q: Does Conveyor see my AI conversation?
A: No. The AI client running on your machine (or in your browser) talks to Conveyor only when it explicitly calls an MCP tool. Your conversation with the AI assistant stays between you and the AI provider.

What's Next?

Connecting Your AI Client — step-by-step setup for Claude, ChatGPT, and API-token clients.
Using the AI Connector — example workflows for security answers, analytics, and knowledge management.
Generating an API Token — create a token and scope its permissions for Bearer-based clients.

Need help? Visit the Integration Issues guide or contact [email protected].