Overview
MCP (Model Context Protocol) Server Connection enables your Utari agents to connect with external tools and services through a standardized protocol. By connecting to MCP servers via HTTP, you can extend your agent's capabilities with custom integrations, specialized tools, and external data sources—creating unlimited possibilities for customization.
MCP Server Connection Process
1. Setup MCP Server
What You Need:
MCP server endpoint URL
Memorable name for the connection
Server must be accessible via HTTP
Process:
1. Get your MCP server URL
Example: https://mcp.yourservice.com/api 2. Choose a descriptive name
Example: "Customer Data API" or "Internal CRM Tools" 3. Enter connection details in Utari
2. Select Tools
After Connection:
Browse available tools from the MCP server
Review tool descriptions and capabilities
Select which tools to enable for your agent
Configure tool permissions and access
Example Tools Discovered:
From "Customer Data API" MCP Server:
✓ get_customer_profile - Retrieve customer information
✓ update_customer_data - Modify customer records
✓ search_customers - Query customer database
✓ get_purchase_history - View transaction history
✗ delete_customer - (Not enabled for safety)
3. Test Connection
Verify Setup:
Test with a simple tool call
Confirm data is returned correctly
Check permissions are working
Validate agent can access tools
4. Use in Workflows
Integration Complete:
Tools now available to agent
Use in conversations naturally
Combine with other Utari tools
Build automated workflows
ℹ️ Note
MCP provides a standardized way for AI applications to securely connect to external tools and data sources, making it easy to extend functionality without custom development.
What is MCP?
Model Context Protocol Explained
Understanding MCP
MCP (Model Context Protocol) is a universal standard that allows AI agents to connect with external tools and services.
Think of it as:
USB for AI tools - standardized connection
Plugin system for agents
API gateway specifically designed for AI
Key Benefits:
✅ Standardized: Works across different AI platforms
✅ Secure: Built-in authentication and access control
✅ Flexible: Connect to any HTTP-accessible service
✅ Discoverable: Tools self-describe their capabilities
✅ Composable: Combine multiple MCP servers
How It Works:
Your Agent → MCP Protocol → MCP Server → Your Service/API 1. Agent requests available tools from MCP server 2. MCP server describes its capabilities 3. Agent calls tools through standardized interface 4. MCP server executes and returns results 5. Agent uses results in conversation/workflow
Setting Up MCP Server Connection
Step 1: Prepare Your MCP Server
Using Existing Server
If you have an MCP server URL:
Information Needed:
1. Server endpoint URL
Example: https://api.company.com/mcp 2. Authentication method (if required)
- API key
- OAuth token
- Basic auth
- None (public server) 3. Any required headers or configuration Ready to connect!
Composio MCP Servers
Using pre-built Composio servers:
Browse available integrations:
- Slack, Gmail, Google Drive
- Salesforce, HubSpot
- GitHub, Jira, Linear
- 100+ other services Process:
1. Search for service in MCP Server Search
2. Get MCP server endpoint
3. Create credential profile
4. Connect via HTTP endpoint
See MCP Server Search for details.
Custom MCP Server
Hosting your own MCP server:
Requirements:
- HTTP-accessible endpoint
- Implements MCP protocol specification
- Returns tool definitions
- Handles tool execution requests Implementation:
- Use MCP SDK for your language
- Define your tools and functions
- Host on accessible infrastructure
- Provide HTTP endpoint URL Resources:
- MCP Protocol Specification
- MCP Server SDKs
- Example implementations
Step 2: Connect in Utari
1. Initiate Connection
Ask your worker:
"Connect to MCP server at https://mcp.myservice.com/api" or "Set up MCP server connection for our internal tools"
2. Provide Details
Worker will ask for: 1. Server URL (required)
"Enter the complete URL to your MCP server endpoint"
→ https://mcp.myservice.com/api 2. Server Name (required)
"Give this MCP server a memorable name"
→ "Internal Customer Tools" 3. Authentication (if needed)
- API key
- Bearer token
- Custom headers
3. Test Connection
Worker attempts to connect: Connecting to https://mcp.myservice.com/api...
✓ Connection successful
✓ Server responded
✓ MCP protocol version compatible Discovering available tools...
4. Review Available Tools
Found 8 tools from "Internal Customer Tools": 1. get_customer_info
Description: Retrieve customer profile data
Parameters: customer_id (required) 2. update_customer_email
Description: Change customer email address
Parameters: customer_id, new_email 3. get_order_history
Description: Fetch customer purchase history
Parameters: customer_id, date_range (optional) [...] Which tools should I enable?
5. Select Tools
You: "Enable get_customer_info, get_order_history,
and search_customers. Don't enable any update or
delete functions" Worker:
✓ Enabled: get_customer_info
✓ Enabled: get_order_history
✓ Enabled: search_customers
✗ Disabled: update_customer_email
✗ Disabled: delete_customer MCP server configured successfully!
Using Connected MCP Servers
Natural Tool Usage
Once connected, agents use MCP tools naturally in conversation:
Direct Requests
"Look up customer ID 12345 using the customer tools""Get the order history for [email protected]""Search for customers in the San Francisco area"
Workflow Integration
"For each lead in the spreadsheet: 1. Look up their customer profile 2. Check their order history 3. Determine if they're a good fit for upsell 4. Create personalized outreach email"
Combined with Other Tools
"Use web search to research the company, then use our customer tools to check if they're already in our system, then draft a personalized email using both sources of information"
Example MCP Server Workflows
Customer Support
Scenario: Customer support agent with internal tools
MCP Server: "Support Tools API" Connected Tools:
- get_customer_info
- get_ticket_history
- get_product_details
- check_warranty_status
- get_return_policy Workflow:
User: "Customer asks about returning product" Agent:
1. Looks up customer info via MCP
2. Checks their order history
3. Verifies warranty status
4. References return policy
5. Provides personalized answer with specifics Result: Fast, accurate, context-aware support
Sales Intelligence
Scenario: Sales agent with CRM and enrichment tools
MCP Servers Connected:
- "Internal CRM" - Customer data
- "Clearbit API" - Company enrichment
- "LinkedIn Tools" - Professional data Workflow:
User: "Research and qualify this lead" Agent:
1. Check if lead exists in CRM (Internal CRM MCP)
2. Enrich company data (Clearbit MCP)
3. Research decision makers (LinkedIn MCP)
4. Synthesize findings
5. Provide qualification recommendation Result: Comprehensive lead intelligence from multiple sources
Data Analysis
Scenario: Analytics agent with database access
MCP Server: "Analytics Database API" Connected Tools:
- query_sales_data
- get_user_metrics
- calculate_conversion_rates
- get_cohort_analysis
- export_report_data Workflow:
User: "Analyze Q4 performance" Agent:
1. Queries sales data for Q4
2. Calculates key metrics
3. Compares to previous quarters
4. Identifies trends
5. Generates report with insights Result: Data-driven analysis with actual company data
Internal Operations
Scenario: Operations agent with multiple systems
MCP Servers:
- "JIRA API" - Project management
- "Slack API" - Team communication
- "GitHub API" - Code repository
- "PagerDuty API" - Incident management Workflow:
User: "Create post-incident report" Agent:
1. Gets incident details (PagerDuty MCP)
2. Finds related tickets (JIRA MCP)
3. Reviews code changes (GitHub MCP)
4. Compiles timeline
5. Generates report
6. Posts to team channel (Slack MCP) Result: Automated cross-system workflow
MCP Server Management
Managing Multiple MCP Servers
1. Connect Multiple Servers
You can connect multiple MCP servers: 1. "Customer Database Tools" (internal)
2. "Payment Processing API" (Stripe)
3. "Analytics Platform" (internal)
4. "Communication Tools" (Slack, Email) Each provides different capabilities
2. Organize by Purpose
Group servers logically: Customer Data:
- CRM MCP server
- Support tools MCP
- Analytics MCP Communication:
- Slack MCP
- Email MCP
- SMS MCP External Services:
- Payment MCP
- Shipping MCP
- Verification MCP
3. Control Access
Different agents get different MCP access: Sales Agent:
✓ CRM MCP (read/write)
✓ Email MCP (send only)
✗ Payment MCP (no access) Support Agent:
✓ CRM MCP (read only)
✓ Ticket MCP (read/write)
✓ Product MCP (read only) Admin Agent:
✓ All MCP servers (full access)
Updating MCP Connections
Update Server URL
If your MCP server endpoint changes: 1. Update the connection URL
2. Retest the connection
3. Verify tools still work
4. Update agent configurations if needed Example:
Old: https://api-v1.company.com/mcp
New: https://api-v2.company.com/mcp
Add/Remove Tools
As your MCP server adds new tools: 1. Refresh connection to discover new tools
2. Review new tool descriptions
3. Enable tools that are relevant
4. Update agent instructions if needed When tools are deprecated:
1. Disable unused tools
2. Update workflows using old tools
3. Test agent still works correctly
Update Authentication
When credentials change: 1. Update API keys/tokens
2. Test connection with new auth
3. Verify all enabled tools work
4. Monitor for auth errors Best practice: Rotate credentials regularly
Disconnect Server
To remove an MCP server: 1. Verify no agents are actively using it
2. Document which agents had access
3. Disconnect the server
4. Update agent configurations Warning: This removes all tools from that server
Security Best Practices
Secure Endpoints
Only connect to MCP servers you control or trust completely
Least Privilege
Enable only the tools agents actually need to use
Authentication
Always use authentication when available (API keys, tokens)
HTTPS Only
Only connect to MCP servers using HTTPS, never HTTP
Monitor Access
Track which agents use which MCP tools and how often
Rotate Credentials
Regularly update API keys and authentication tokens
Audit Logs
Review MCP server logs for unusual access patterns
Test Safely
Test MCP connections with non-production data first
Troubleshooting
Connection fails
Check:
Server URL is correct and accessible
Server is currently running
No firewall blocking the connection
HTTPS certificate is valid
Authentication credentials are correct
MCP server implements protocol correctly
No tools discovered
Verify:
MCP server is returning tool definitions
Server implements MCP protocol correctly
API endpoint is the correct path
Authentication is working
Check server logs for errors
Tool calls fail
Debug:
Verify tool is enabled for agent
Check parameters are correct format
Confirm authentication is still valid
Review MCP server logs
Test tool manually if possible
Check for rate limiting
Slow response times
Consider:
MCP server performance/load
Network latency to server
Complex operations taking time
Rate limiting causing delays
Optimize tool calls if possible
Cache frequently accessed data
Authentication errors
Resolve:
Verify credentials are current
Check if tokens expired
Confirm API key has required permissions
Ensure auth headers formatted correctly
Rotate credentials if compromised
Contact server admin if persistent
Summary
You've successfully learned how to:
✅ Success
Understand what MCP servers are and how they work
✅ Success
Connect to MCP servers via HTTP endpoints
✅ Success
Select and enable specific tools from MCP servers
✅ Success
Use MCP tools naturally in agent workflows
✅ Success
Manage multiple MCP server connections
✅ Success
Apply security best practices for MCP connections
✅ Success
Troubleshoot common connection and tool issues
MCP Server Connection provides unlimited extensibility for your Utari agents, allowing them to connect with any HTTP-accessible service through a standardized protocol—transforming agents from general-purpose assistants into specialized tools with deep integration into your systems and workflows.
Next Steps
Discover pre-built MCP servers for popular services
Manage authentication for MCP server connections
Configure agents with MCP server access
Create specialized agents with MCP integrations