MCP Servers
Connect external tools and services to AI SmartTalk using the Model Context Protocol (MCP). MCP is an open standard that lets your AI assistant interact with external systems through a unified interface.
Overview
The MCP Server integration enables you to:
- Connect multiple MCP servers simultaneously
- Expose external tools for your AI to call during conversations
- Authenticate with custom HTTP headers (API keys, tokens, etc.)
- Test connections before going live
- Extend capabilities without custom development
What is MCP? Model Context Protocol is an open standard (developed by Anthropic) that allows AI systems to securely connect to external data sources and tools. Learn more at modelcontextprotocol.io.
Prerequisites
Before you begin, ensure you have:
- An active AI SmartTalk account
- Access to one or more MCP-compatible servers
- The server URL for each MCP server
- Any required authentication credentials (API keys, tokens)
How MCP Works
- User asks a question that requires external data
- AI SmartTalk recognizes it needs an MCP tool
- The appropriate MCP server is called with the right tool
- Response is returned and the AI formulates a natural language answer
AI SmartTalk supports multiple MCP servers at once. Tools from different servers are automatically namespaced to prevent conflicts.
Step-by-Step Setup
Step 1: Access MCP Integration
- Log into your AI SmartTalk account
- Navigate to Administration → My Assistants → [Your Assistant Name]
- Click on Connectors in the left sidebar
- Locate MCP Servers and click on it
Step 2: Add an MCP Server
Click Add Server to open the server configuration form. Fill in the following fields:
| Field | Required | Description |
|---|---|---|
| Server Name | Yes | A friendly name to identify this server (e.g., "Company CRM", "Internal Wiki") |
| Server URL | Yes | The MCP server endpoint URL (e.g., https://mcp.example.com) |
| HTTP Headers | No | Custom headers for authentication or other requirements |
AI SmartTalk automatically appends /mcp to your URL if it doesn't already end with it. For example, https://example.com becomes https://example.com/mcp. If your server uses a different endpoint path, include it in the URL (e.g., https://example.com/api/mcp).
Step 3: Add Authentication Headers (Optional)
If your MCP server requires authentication, click Add Headers to configure custom HTTP headers:
| Common Header | Use Case |
|---|---|
Authorization | Bearer tokens (e.g., Bearer sk-xxx...) |
X-API-Key | API key authentication |
X-Auth-Token | Custom token authentication |
For each header, provide:
- Key: The header name (select from common headers or type a custom one)
- Value: The header value (sensitive values are automatically masked in the UI)
You can add multiple headers per server. Headers containing "secret", "password", "token", or "auth" in the key name are automatically hidden for security.
Step 4: Test the Connection
Before saving, click Test Connection to verify your server is reachable:
- AI SmartTalk sends a JSON-RPC initialization request to your server
- The test checks multiple endpoints in order:
/mcp,/health, then the base URL - Results show connection status, response time, and server information
| Test Result | Meaning |
|---|---|
| Connected | Server responded successfully with MCP protocol |
| Authentication failed | Check your API key or token in headers |
| Server not found | Verify the URL is correct |
| Connection refused | Server may be down or firewall blocking access |
| Timeout | Server took too long to respond (10s limit) |
| SSL error | Check that the server has a valid SSL certificate |
Step 5: Save Configuration
- Click Add to add the server to your list
- Repeat Steps 2-4 to add more MCP servers if needed
- Click Save Integration to activate all configured servers
Multi-Server Support
AI SmartTalk supports connecting multiple MCP servers simultaneously. This allows you to combine tools from different sources:
Tool Namespacing
Tools from each server are automatically prefixed with the server name to prevent naming conflicts. For example:
company_crm__search_customersinternal_wiki__search_docsweather_api__get_forecast
This means you can safely connect servers that expose tools with the same name.
Managing Multiple Servers
The integration dashboard shows all configured servers with:
- Server name and URL for easy identification
- Headers count badge if custom headers are configured
- Status indicator (Configured / Incomplete)
- Action buttons: Test connection, Edit, or Delete individual servers
Tool Discovery
Once your MCP servers are connected, AI SmartTalk automatically discovers all available tools exposed by each server. These tools become available to your AI assistant immediately.
The AI assistant will:
- See the list of all available tools from all connected servers
- Automatically select the right tool based on user queries
- Call the tool with appropriate parameters
- Process the response and provide a natural language answer
You don't need to manually configure or enable individual tools. All tools exposed by a connected MCP server are automatically available to your AI assistant.
Use Cases
Customer Relationship Management
Connect your CRM via MCP:
- "Look up customer John Smith"
- "What's the status of account #12345?"
- "How many open support tickets do we have?"
Internal Tools
Access company systems:
- "Check inventory for product SKU-123"
- "What's on the schedule for tomorrow?"
- "Search our internal wiki for onboarding docs"
External Data Sources
Fetch live data:
- "What's the current weather in Paris?"
- "Get the latest stock price for AAPL"
- "Check flight status for UA123"
Development Tools
For technical teams:
- "What's the status of the latest build?"
- "Search our GitHub issues for authentication bugs"
- "Check the health of production servers"
MCP Server Examples
Popular MCP Servers
| Server | Purpose | Link |
|---|---|---|
| Filesystem | Read/write local files | Built-in reference server |
| PostgreSQL | Query databases | Community server |
| GitHub | Repo management | Community server |
| Slack | Messaging integration | Community server |
| Google Drive | File management | Community server |
| Brave Search | Web search | Community server |
Find more: Visit the MCP Server Registry for a growing list of community MCP servers.
Self-Hosting MCP Servers
You can create custom MCP servers for:
- Proprietary internal systems
- Custom business logic
- Secure or private data access
Use any MCP SDK (TypeScript, Python) to build your own server.
Security Considerations
Authentication
Configure authentication via HTTP Headers in the server configuration:
| Method | Header Example |
|---|---|
| Bearer Token | Authorization: Bearer sk-xxx... |
| API Key | X-API-Key: your-api-key |
| Custom Token | X-Auth-Token: your-token |
Data Privacy
- MCP requests may contain user conversation data
- Ensure your MCP server handles data according to your privacy policies
- Consider data residency requirements
- AI SmartTalk masks sensitive header values in the UI
Network Security
- Use HTTPS for all MCP connections
- Restrict MCP server access to trusted IPs
- Implement rate limiting on your MCP server
- Ensure valid SSL certificates are in place
Troubleshooting
Connection Issues
| Issue | Solution |
|---|---|
| "Connection refused" | Check server URL and port. Ensure the server is running |
| "Authentication failed" (401) | Verify your API key or token in the headers configuration |
| "Forbidden" (403) | Check that your credentials have the required permissions |
| "Server not found" (404) | Verify the URL. AI SmartTalk appends /mcp automatically |
| "Timeout" | Check network connectivity and firewall rules. The test has a 10-second timeout |
| "SSL error" | Ensure your MCP server has a valid SSL certificate |
| "Rate limited" (429) | Too many requests. Wait and try again, or increase server limits |
Tool Issues
| Issue | Solution |
|---|---|
| Tools not appearing | Verify the MCP server exposes tools correctly. Try refreshing |
| Wrong tools called | Check server names are descriptive to help the AI choose correctly |
| Tool call fails | Check server logs for errors. Ensure tool returns expected format |
Common Mistakes
- Forgetting
/mcp: AI SmartTalk adds it automatically, but if your endpoint is different, include the full path - Wrong header format: Make sure authentication headers match what your server expects
- Duplicate server names: Each server must have a unique name
Managing MCP Servers
| Action | How |
|---|---|
| Add server | Click "Add Server", fill in details, test, and save |
| Edit server | Click the edit icon on the server card |
| Test connection | Click the test icon on the server card |
| Delete server | Click the delete icon on the server card |
| Save changes | Click "Save Integration" to persist all changes |
Best Practices
- Use descriptive server names: Helps the AI understand which server to query (e.g., "Company CRM" rather than "Server 1")
- Test before saving: Always test connections to catch configuration issues early
- Start with one server: Get comfortable with one integration before adding more
- Secure your credentials: Use dedicated API keys for AI SmartTalk, not personal tokens
- Monitor your servers: Track which tools are called and their performance
- Keep servers updated: Ensure your MCP servers follow the latest protocol specification