AI Tools let AI Agents and Copilot execute functions, look up data, provide insights, or integrate with external systems—via APIs or Model Context Protocol (MCP). These actions are customizable and prompt-driven, allowing them to be triggered based on natural language inputs. MCP enhances the precision of AI responses, resulting in smarter and more relevant automation. MCP (Model Context Protocol) Server enhances AI accuracy by providing contextual awareness. It helps the AI agent understand user intent and session details. By leveraging this context, MCP ensures that AI-driven actions are both relevant and precise, improving the overall support experience.

Why use MCP in BoldDesk?

• Contextual awareness: Provides intent and session state to improve tool selection and answers.
• Composable actions: Consolidates multiple external functions (refunds, lookups, updates) behind one protocol.
• Natural‑language triggers: Tools run from prompts when inputs are satisfied.

Steps to Configure MCP Server

Follow the steps below to configure MCP Server.

1. Navigate to MCP

• Navigate to AI module > AI Tools > MCP.
  
  
  
  
  
• Here, you can view existing actions and enable or disable them based on your requirements.

2. Create MCP Server

• Click the Create MCP button to add a new AI Action.
  
  
  
  
  
• Configure the following settings:
  - Applied To: Specify whether the AI Action is to apply to AI Copilot, AI Agent, or both.
  - Brand: Select which brand(s) the MCP Server applies to.
  - Name: Set the name of the MCP Server.
  - Usage Criteria: Set the usage criteria.
  - MCP Endpoint: Connect and manage external model actions through a secure MCP endpoint. Enable your AI Agents to trigger remote workflows with real-time data exchange and control.
  - Server URL: Provide the server URL. This URL allows BoldDesk to connect to a Model Context Protocol (MCP) server so the AI can use the server’s tools, context, and capabilities.
  - Headers: Specify the API endpoint that should be called to access data or be sent updates. You can pass AI Agent attributes in the URL or request body.
• Click on next.

3. List Tools

• Use List Tools to fetch available tools from the MCP server.
• Internally, this calls the MCP operation tools/list to return tool names, descriptions, and input schemas.
• After publishing, use Check new tools to refresh when the server adds or changes tools.
  Example outcome:
  • get_order_status(order_id)
  • process_refund(order_id, reason)
  • update_profile(email, phone)

4. Publish or Save as Draft

• Choose Publish to go live or Save as Draft to continue editing later. You can also use the preview option to test the MCP server.
  
  
  
  
  

Usage Examples

Example A — Order Lookup

• User prompt: “Track order #A123.”
• Tool used: get_order_status(order_id)
• AI behavior: Extracts A123, calls the tool, returns carrier, ETA, and status.

Example B — Refund Processing (Guardrails)

• User prompt: “Refund order A123 for damaged item.”
• Tool used: process_refund(order_id, reason)
• Considerations: Require agent review or role‑based approval, handle partial refunds, capture reason codes.

Example C — Profile Update

• User prompt: “Change my phone number to +1‑555‑1234.”
• Tool used: update_profile(email, phone)
• AI behavior: Confirms identity, validates format, updates record, logs change.

Edge Cases & Troubleshooting

• Error: “Invalid MCP credentials: no matching server found.”

  • Cause: The Server URL does not resolve to an MCP‑compliant server, incorrect path, or auth missing.
  • Fix: Verify the server is live, confirm the endpoint path, add required headers, and re‑run List Tools.

• Tools not appearing after publish.

  • Cause: MCP server added tools after initial sync or returned an empty list.
  • Fix: Click Check new tools, verify tools/list returns tools with proper schemas.

• 401/403 authorization failures.

  • Cause: Missing or expired token.
  • Fix: Update Headers (e.g., Authorization: Bearer <token>), confirm token scope.

• Schema mismatch or invalid inputs.

  • Cause: Tool requires fields not collected in conversation.
  • Fix: Adjust prompts or MCP input schema; enforce Usage Criteria to request necessary data.

• Environment confusion (staging vs production).

  • Cause: Wrong Server URL or keys.
  • Fix: Use environment‑specific endpoints and rotate credentials per environment.

Permission Required to Manage AI Tools

To manage AI Tool, users must have Manage AI Actions permission.

Frequently Asked Questions (FAQs)

Q: How is an MCP Server different from API Actions?
A: API Actions call a single external endpoint per action you configure in BoldDesk. An MCP Server centralizes multiple actions (“tools”) and provides context (intent/session) that helps AI pick and execute the right tool from natural‑language prompts.

Q: Do I need separate MCP servers for each brand?
A: Not required. You can bind one MCP Server to multiple Brand(s), but ensure authorization and data segregation are enforced by the server.

Q: What if my server adds a new tool later?
A: Click Check new tools after publish to resync. Ensure tools/list returns the new tool with a valid input schema.

Q: Can I pass ticket or requester attributes to the server?
A: Yes. Include them as Headers, query params, or body fields—aligning with your MCP server’s contract and privacy rules.

Related Articles

• How to Create and Use Custom Prompts
• Get Started with BoldDesk AI Chatbots