Skip to main content

​
What is MCP?

Model Context Protocol (MCP) is a standardized way for AI agents to access external tools and services. Velatir’s MCP server provides a security-focused approval gate that requires human authorization before AI agents can execute potentially sensitive operations.

​
Key Features

Human-in-the-Loop Approval

Mandatory approval gate for sensitive AI agent operations

Request Tracking

Unique ID-based approval system with real-time status polling

Feedback Integration

Denial reasons for iterative improvement of AI agent behavior

Universal Compatibility

Works with any MCP-compatible AI tool or platform

​
Supported AI Tools & Platforms

​
AI Assistants

Claude Desktop - Native MCP support for human approval workflows
Claude Web - Via MCP protocol for controlled AI interactions
Claude API - Integrate approval gates in custom Claude implementations
ChatGPT Desktop - MCP integration for approval workflows
Custom GPT Apps - Add human oversight to specialized GPT applications
OpenAI API - Embed approval gates in OpenAI-powered tools
Ollama - Human approval for local model operations
LM Studio - Approval gates for locally hosted AI models
Text Generation WebUI - Integrate with web-based AI interfaces

​
Development Tools

Cursor - Approve AI code changes before execution
GitHub Copilot - Human review for AI-generated code
Replit - Approval gates for AI-assisted development
Windsurf - Human oversight for AI development workflows
LangChain - Native middleware integration available (see LangChain Integration) CrewAI - Human oversight for multi-agent systems AutoGPT - Approval gates for autonomous AI operations Semantic Kernel - Microsoft’s AI framework integration

​
Business & Productivity Tools

Zapier - Human approval for AI-powered automations
n8n - Approval nodes in workflow automation (also has dedicated node)
Microsoft Power Automate - AI approval gates in business workflows
Integromat/Make - Human oversight for automated processes
Tableau - Approve AI insights before sharing
Power BI - Human review of AI-generated reports
Custom BI Tools - Integrate approval into data analysis workflows

​
Agent Frameworks

Microsoft Autogen - Human oversight for agent conversations
Google AI Studio - Approval gates for AI experiments
Hugging Face Agents - Human review for model interactions
Custom Agent Platforms - Universal MCP compatibility

​
Available Tools

​
request_human_approval

Initiates a human approval request before sensitive operations. Parameters:
  • functionName: Name of the function requiring approval
  • args: Arguments or context for the action
  • description: Human-readable description
  • metadata: Additional metadata for context
Returns: Approval request ID for status tracking

​
check_approval_status

Checks the current status of an approval request. Parameters:
  • requestId: The ID returned from request_human_approval
Returns: Current status (pending, approved, denied, changerequested) and optional feedback

​
Response Status Types

Request Submitted - Waiting for human review
Next Action: Continue polling until status changes
Request Approved - Human reviewer authorized the operation
Next Action: Proceed with the originally requested operation
Request Denied - Human reviewer rejected the operation
Next Action: Do not proceed; check feedback for reasons
Changes Requested - Human reviewer wants modifications before approval
Next Action: Review feedback, make requested changes, and submit a new approval request with the updated parameters

​
Handling Change Requests

When a request receives changerequested status, the AI agent should:
  1. Parse the feedback to understand what changes are needed
  2. Modify the request based on human feedback
  3. Submit a new approval request with updated parameters
  4. Reference the original request in metadata for audit trail

​
Security Considerations

The Velatir MCP server is designed as a defensive security tool:
  • Approval Gates - Prevents unauthorized operations by requiring explicit human consent
  • Audit Trail - Every request and decision is logged for compliance
  • Secure Communication - All communications are encrypted and authenticated
  • No Execution - The server never executes operations, only manages approvals

API Integration

Direct REST API integration

n8n Node

Workflow automation integration