Quick Start
Updated on 3 November 2026.
Add an MCP server, choose what agents may call, connect an AI app and review the first tool calls.
Video walkthrough
Registering servers, governing which tools agents can call, reading the audit trail, and handing the gateway to your users.
YouTube All videosThe rest of the series: the user dashboard for the people who will use the MCPs, and local setup for administrators and for users.
1. Add an MCP
Go to Settings > AI Gateway > MCP Gateway.
If the server uses OAuth, a banner asks for an administrator connection. Click Review all (or Connect OAuth on the card) and sign in once so the gateway can discover its tools.
2. Configure
After you add a server, review who can connect and which tools they may call before connecting users. Choose the mode that governs your MCP Gateway.
- Classic settings
- Policy Engine
With the MCP Gateway Policy Engine off, configure the server directly. New servers' tools stay off until you enable them.
Click Save settings before you close the drawer. For every section, see the Overview.
With the engine on, open Govern > Policy Engine > MCP Gateway and review the matching rules:
- MCP Server Access decides who may connect.
- Invocation & Discovery decides which tools are listed, may run and may return results. Select a recipe such as Allow selected tools and choose the server and tools.
- Data & Adversarial Risks selects checks and actions for tool arguments and results. Use Human Confirmation where a caller must approve a tool call.
- Save changes to the shared draft, use Test, then Review & publish.
Check existing matching rules before adding new ones. Published policies govern the marked settings sections; editing their classic values does not change live policy behavior. See Author, simulate and publish and Tool visibility.
3. Connect your AI app
Users sign in at mcpgateway.quilr.ai, connect their accounts and copy a URL from there.
The base domain can differ by environment. Copy the exact URL from the server card or the user dashboard.
OAuth-capable clients such as Claude and ChatGPT sign in when they first connect. For a client that sends a token instead, create one in the server's Settings > API tokens (see API Tokens) and send it with the user's email:
{
"mcpServers": {
"quilr-mcp": {
"url": "https://mcpgateway.quilr.ai/<your-mcp-slug>/mcp",
"headers": {
"Authorization": "Bearer <your-token>",
"mcpuser": "user@company.com"
}
}
}
}
See the Integration Guide for more client examples.
4. Monitor
Select Overall analytics to open MCP Gateway Insights.
For an individual call, return to a server card and select Inspect > Logs. The operational Activity > Tool calls view shows its user, agent and outcome. Select a call for its details. See Analytics and insights for the full tour.
To send logs to your own data platform, use the Log Export API.
Related
- Overview - the MCP Gateway page, server card and workspace.
- Analytics and insights - adoption, protection, reliability and tool savings.
- Adding MCP Servers - every way to add a server.
- Agent Custom Instructions - configure clients to use the OneMCP discovery flow.