Skip to main content

MCP Tool Usage Instructions

AI MCP lets you connect Brightcall.ai's actions with any AI tool that supports MCP

Updated over a week ago

Overview

To ensure that MCP (Model Context Protocol) tools are called correctly by the AI assistant, you must explicitly mention the tool in your prompts. Without explicit instructions, the probability of the tool being invoked decreases significantly.

Add input with name mcpServerUrl with value your MCP server url.

IMPORTANT: currently we don’t support header API key authorisation.

Required Format

When you need the AI assistant to use an MCP tool, include the following instruction in your prompt:

  • If you need to send data to CRM use tool brightcallmcp_toolname.

  • Where toolname is the actual name of the tool as defined in the MCP server.

Example

Here is a real example from a working setup: If you need to send data to CRM use tool brightcallmcp_google_calendar_api_request_beta.

Important Notes

  1. Explicit Instructions Required: The AI assistant needs explicit instructions to call MCP tools. Simply having the tool available is not sufficient.

  2. Tool Naming: The tool name must match exactly as it appears in your MCP server configuration, prefixed with brightcallmcp_.

  3. Placement in Prompts: Include this instruction in the relevant section of your prompt where the tool should be used (e.g., in the knowledge base, campaign data, or assistant instructions).

  4. Context Matters: Provide context about when and why the tool should be used to improve the likelihood of correct invocation.

Best Practices

  • Include the tool instruction in the section of your prompt that describes the action requiring the tool

  • Be specific about when the tool should be used

  • Test the prompt multiple times to ensure the tool is being called reliably

  • If the tool is not being called, try making the instruction more explicit or adding it to multiple relevant sections of your prompt

Troubleshooting

  1. Verify the tool name matches exactly (including the brightcallmcp_ prefix)

  2. Ensure the instruction is explicitly stated in the prompt

  3. Check that the instruction is placed in a relevant context within the prompt

  4. Consider adding the instruction to multiple sections (knowledge base, campaign data, etc.)

  5. Test with different phrasings if the current instruction is not working

Did this answer your question?