Skip to main content

Agents & MCP Settings

This section manages your agents, MCP server connections, and conversation rules.

Agents tab​

Lists all available agents — both built-in and custom.

For each agent, you can:

  • Enable / Disable the agent with a toggle
  • Edit the agent's configuration
  • Delete custom agents (built-in agents can only be disabled)

Creating a custom agent​

  1. Click New Agent
  2. Fill in:
    • Name — Display name shown in the UI
    • Description — Explains what the agent does (helps you remember its purpose)
    • System Prompt — The instruction text the model receives when this agent is active
    • Temperature — Controls response creativity (0 = precise, 1 = creative)
    • Max Tokens — Maximum length of responses
    • Tool Definitions — Custom tools this agent can call (JSON format)
  3. Save the agent

Import / Export agents​

  • Import — Load agents from a .json file (useful for sharing or restoring agents)
  • Export — Save all custom agents to a .json file

MCPs tab​

Lists all configured MCP (Model Context Protocol) servers.

Each server shows its connection status and supports:

  • Connect — Start the connection
  • Disconnect — Stop the connection
  • Reconnect — Disconnect and reconnect
  • Edit — Change server configuration
  • Delete — Remove the server

Adding a new MCP server​

  1. Click Add Server
  2. Fill in the configuration:
    • Name — A label for this server
    • Type — Local or Remote
    • Command / URL — How to reach the server
    • Environment variables — Key=value pairs (optional)
    • Auto-connect — Connect automatically on startup
  3. Click Test to verify the connection
  4. Click Save and Connect

For full details, see MCP Servers.

Rules tab​

Rules are persistent instructions that are injected into the system prompt across all conversations (unless disabled).

Creating a rule​

  1. Click New Rule
  2. Fill in:
    • Name — A label to identify the rule
    • Type — Choose from:
      • System — Added to the system prompt
      • Behavior — Modifies the model's response style
      • Context — Provides background context about you or your work
    • Text — The instruction text
  3. Save

Managing rules​

  • Enable / Disable with a toggle per rule
  • Edit to modify content
  • Delete to remove permanently

Example rules​

Example 1 — Language preference:

  • Name: Language
  • Type: System
  • Text: Always respond in British English.

Example 2 — Code style:

  • Name: Code style
  • Type: Behavior
  • Text: When writing code, always use TypeScript, prefer functional patterns, and add JSDoc comments to all public functions.

Example 3 — Background context:

  • Name: My role
  • Type: Context
  • Text: I'm a backend engineer at a fintech startup. Our stack is Node.js, PostgreSQL, and Kubernetes. We follow strict SOC 2 compliance requirements.