Screens / Knowledge and setup
Connecting an MCP server
Let a person decide whether an agent may use an MCP server: where it runs, who runs it and what it can do.
For: Admins and agent owners adding a tool source.
3 ways to lay it out
1. Consent dialog
Pick it when: The default: a short, deliberate step before anything connects.
- MCP connectProConsent to connect: where it runs, who runs it, what it says it can do (its tools, by name).
- Copy fieldFreeA value to copy, such as a callback URL.
- LockedFreeConnect stays locked, with the reason, until the person has seen the tools.
Pro and Premium items need a licence key; free ones install as they are.
2. Connections page
Pick it when: Many servers to see, review and remove.
- Settings layoutProSections down the side.
- Tool permissionsProWhat each of the server's tools may do.
- MCP inputFreeThe input a tool takes, shown as it is.
Pro and Premium items need a licence key; free ones install as they are.
3. A server's own UI in the thread
Pick it when: The server ships an MCP App and the person uses it in the conversation.
- MCP app frameProAn MCP App isolated in a sandbox on another origin; without a sandbox it says why instead.
- ThreadFreeThe conversation around it.
- Composer sourcesProThe connected servers in the composer's @ menu, each saying whether it is connected, needs signing in or can't be used, and why.
Pro and Premium items need a licence key; free ones install as they are.
Rules that matter most here
- Unknown stays unknown
A server's own description of its tools is shown as the server's claim, not as verified.
- A control that can't be used says why
Connect says why it's not yet available, for example that the tools haven't been shown.
- Say what can and can't be undone
Removing a server says which agents lose which tools.
Brief for a coding agent
Paste this into the agent that builds the screen. The docs MCP server returns the same text from get_screen.
Build the connect-an-MCP-server step with Candor UI: McpConnect asks for consent, naming where the server runs, who runs it and what it says it can do (its tools, by name). Show the tool list as the server's own claim; Candor does not verify it. Connect stays locked (Locked) until the tools have been shown, and says why. After connecting, list the server's tools in ToolPermissions; new tools start as ask first. A server's own UI (an MCP App) goes in McpAppFrame, isolated in a sandbox on another origin; without a sandbox, the frame says why it isn't shown. Removing a server says which agents lose which tools. Start from the "Consent dialog" layout: A dialog listing the server's host, its owner and its tools by name, with Connect and Cancel. Candor items: mcp-connect (https://candor.themethodic.com/docs/mcp-connect.md), copy-field (https://candor.themethodic.com/docs/copy-field.md), locked (https://candor.themethodic.com/docs/locked.md). Install: npx candor-ui add mcp-connect copy-field locked Read each item's docs before using it and take its props from there; don't guess them. Pro and Premium items need the person's licence key, which they add with `npx candor-ui key <key>`; don't ask for it.