Creating Your First Agent

Creating an AI Agent in Chase Agents takes under five minutes. This guide walks you through the process from start to first conversation, including how to write a system prompt that makes your agent genuinely useful.

Method 1: Using the Dashboard

From your workspace dashboard, navigate to the Agents section and click New Agent. You will be prompted to provide a title, a system prompt, and optionally a description. After saving, you can assign MCP connections to the agent from the connections panel.

Method 2: Using the MCP create_agent Tool

If you are working within Claude Desktop or another MCP-connected AI client, you can create agents programmatically using the create_agent tool. Call it with a title, a prompt (the system prompt), and a description. The tool returns the new agent's ID which you can use immediately to start a session.

Writing an Effective System Prompt

The system prompt is the most important part of your agent. It shapes every response the agent gives and determines how it uses its tools. A well-written system prompt states the agent's role and purpose clearly in the first sentence, lists the connections the agent has access to and explains what each one is for, describes the kinds of tasks the agent should help with and those it should decline, specifies the preferred output format such as concise bullet points or markdown tables, and mentions any important constraints such as never modifying production records without confirmation.

Assigning Connections (Server Mappings)

After creating the agent, you assign which MCP connections it can use. Each connection exposes a set of tools. For example, a Slack connection exposes tools for reading channels, posting messages, and listing members. The agent can only call tools from connections explicitly assigned to it.

Follow the principle of least privilege: give the agent access only to the connections it genuinely needs. A customer-support agent does not need access to your code deployment tools. An analytics agent does not need write access to your CRM. Narrowing scope reduces risk and makes the agent's behavior more predictable.

Starting a Chat Session

Once your agent is created and connections are assigned, click Start Chat to open a new session. The agent receives your first message alongside its system prompt and the full list of available tools. It begins reasoning immediately. You can ask it anything within its scope and it will call tools, observe results, and respond in natural language.

Connection Access Types

When assigning a connection to an agent, you choose the access type. Personal connections are tied to your individual credentials. WorkspaceWide connections are shared across the workspace and available to all agents by default. SpecificTeamMembers connections can be scoped to a subset of users. Understanding these access types helps you ensure the agent uses the right authentication context for each operation.

Iterating on Your Agent

Agents improve with iteration. After your first few conversations, review where the agent gave unhelpful responses and refine the system prompt to address those gaps. Common adjustments include clarifying what data sources to prioritize, specifying response length preferences, and adding examples of the ideal output format. Small prompt changes often produce significant improvements in quality.

Sharing Agents with Your Team

Agents you create in your workspace are available to all workspace members by default. Team members can open new chat sessions with any agent, each session maintaining its own independent conversation context. Workspace admins can review, edit, or remove agents from the workspace settings panel.