MCP Server

MCP Server

Quickly hosts a Model Context Protocol server, which lets an external AI client discover and call your workspace's tools directly. Where the REST API is for software you write, MCP is for connecting another AI system to Quickly without writing an integration at all.

Who can use it: Owner, Admin · Availability: Core surface

What a Client Gets

A connected client discovers the tools your workspace exposes and can call them as its own. What it can reach is bounded by the same things that bound everything else in Quickly: the OAuth scopes granted during connection, your workspace's tool policies, and the access of the identity behind the connection. Destructive tools stay blocked regardless.

Authorization

Connection runs through the app's OAuth surface rather than a shared secret, and every route enforces its own scope checks. That means a connected client is a revocable grant you can inspect, not a key someone pasted into a config file and forgot about.

Deciding What to Expose

Treat an MCP connection as another consumer with a role. Review the tool catalog before connecting a client, and use workspace tool policies to narrow what is reachable rather than relying on the external client to behave. A tool set to require confirmation still requires it when the call arrives over MCP.

How to connect an MCP client

1. Open Developers in the sidebar and select the MCP Server tab.
2. Copy the server URL.
3. Paste it into the MCP-compatible client and complete authorization.

How to review OAuth setup

1. Check the discovery details on the MCP Server tab.
2. Confirm the client supports OAuth or the required auth method.
3. Reconnect if authorization expires.

How to control external tool access

1. Use workspace roles, scopes, and tool policy settings.
2. Review Action Logs for client activity.
3. Click Revoke if a client should no longer use Quickly tools.

Good to Know

• MCP authorization is handled through the app OAuth surface and route-level scope checks.

Related

• API Keys and Developers for programmatic access over REST instead.
• Autonomy Controls and Safety for the policies that apply to MCP tool calls.
• The MCP server page for setup details and supported clients.