Configure HubSpot's AI agent prompts with connected MCP tools

After connecting an MCP server to a HubSpot's AI agent, configure prompt instructions to determine when and how the agent uses connected MCP tools.

Well-designed prompts help agents combine HubSpot data with information from external systems to complete specific business tasks consistently.

Add MCP prompts to HubSpot's AI agents

1. In your HubSpot account, click More, then navigate to Agents > Agent Hub. If More doesn't appear in your account, navigate to Agents > Agent Hub directly.
2. Click the Agents tab.
3. Hover over the agent you want to edit and click Edit.
4. In the Instructions section, enter your MCP prompts. For example, you can use the following prompt: 
   
   Example prompt: 
   "You are connected to the G2 MCP Server. For every deal analysis, always query the G2 MCP Server to retrieve buyer intent signals, competitor comparison activity, and intent scores for the companies associated with closed-lost deals. Combine G2 external market signals with HubSpot CRM data to provide a complete picture of loss reasons. When a competitor is identified in G2 data, cross-reference it against the deal's loss reason in HubSpot. Surface patterns where G2 intent activity spiked or shifted toward a competitor before deal close."

Use case: create automated account intelligence from sales conversations

The following example can be used as a guide for adding MCP prompts to a HubSpot's AI agent connected to Notion and Gong MCP servers. While this use case offers a general framework, you can customize it for your specific goal and for other MCP servers.

Business objective

A sales organization aims to reduce the time account teams spend researching customer conversations, allowing them to focus more on engaging with prospects.

Historically, reps must manually review call recordings, search for mentions of competitors, and document key takeaways across multiple systems.

By connecting HubSpot's AI agents to external tools through HubSpot MCP Client, the organization can centralize research activities and automatically generate actionable insights.

Example configuration

This example shows how a Sales team connects a HubSpot's AI agent to external systems using HubSpot MCP Client.

1. Connect the agent to Gong MCP server to analyze sales calls and identify recurring competitive objections.
2. Connect the agent to Notion MCP server to store account summaries in a shared workspace.
3. Configure MCP prompts for the agent following the best practices below.

Best practices for MCP prompts

• Identify the database: since Notion pages and databases don't follow a fixed structure, you need to instruct the agent exactly on which database to use.
  • Best practice: you should mention the name of the database, its purpose, and include the database ID when possible.
  • Example prompt: Use the Notion database titled 'Content Projects Tracker,' which includes the project name, status, owner, and due date.
• Describe the property or column structure: describe your database structure in detail to help the agent understand the data.
  • Best practice: when referencing a database, you should list the relevant column names, what they contain, and their property type. If some fields are optional or not consistently used, they should be included.
  • Example prompt: The database includes these columns: 'Project Name' (title), 'Owner' (person), 'Status' (select), and 'Deadline' (date).
• Define the agent’s task: be clear about the specific information the agent should find or extract.
  • Best practice: when defining the agent's task or requesting for data, use specific names and details. 
  • Example prompt: Find all items where the Status is 'In Progress' and the Deadline is within the next 7 days.
• Specify actions and tools: if your agent needs to perform a specific action, explicitly state which tools it should use and how. This is especially useful if the agent has to choose between multiple tools.
  • Best practice: define the exact tool to use and how any actions should be executed, such as query_database, list_databases, or retrieve_page.

• Provide an example input and output: if your agent needs to add or edit entries, it’s recommended that you include an example of what the data should look like.
  • Best practice: provide a clear example of each input and desired output with examples. You can also add conditions for different outputs.
  • Example prompt: When creating a new entry, fill in 'Status' as 'Not Started' by default and leave 'Deadline' blank.

Result

After the configuration is complete:

• HubSpot's AI agents can access data from connected business systems.
• Sales teams can analyze customer conversations without leaving HubSpot.
• Account summaries, competitor insights, and follow-up actions are automatically generated.
• Teams spend less time gathering information and more time taking action.