The Problem (Pain Level: 8/10)

“The AI agent sent emails, deleted files, and processed payments automatically… without any confirmation” - A new nightmare for companies adopting AI agents.

Current pain points:

  • Loss of control: AI agents executing critical operations without approval
  • No audit trail: Can’t trace which tools an agent called and why
  • No rollback: Can’t undo dangerous operations already executed
  • Compliance violations: Automation without human review causes regulatory issues
  • Trust barrier: Psychological reluctance to deploy agents in critical workflows

Target Market

Primary Target: Companies adopting AI agents, SaaS startups, enterprise IT teams

Market Size:

  • AI orchestration market: $30B+ by 2027 (CAGR 65%)
  • 35% of enterprises allocating $5M+ budget for AI agents in 2026
  • MCP SDK monthly downloads at 97M - steep technical adoption curve
  • 25% of AI agent evaluators expected to move into higher-value roles by 2028

Pain Frequency: Control needs increase as agent automation grows

What is Preloop?

A Human-in-the-Loop proxy gateway positioned between AI agents and external tools.

Core Concept:

# Preloop policy configuration example
policies:
  - name: "Email Sending Approval"
    trigger:
      tool: "send_email"
      conditions:
        - recipients: "external"
        - contains_pii: true
    action: "require_approval"
    timeout: "30m"

  - name: "Payment Processing Block"
    trigger:
      tool: "process_payment"
      conditions:
        - amount: ">$100"
    action: "block_with_review"

  - name: "File Deletion Alert"
    trigger:
      tool: "delete_file"
    action: "notify_and_log"

Differentiation:

  • MCP Native: Full Model Context Protocol standard support
  • Policy-Based Control: Define approval conditions with declarative rules
  • Real-Time Dashboard: Pending approvals, executed operations, blocked requests at a glance
  • Audit Logging: Automatic recording and search of all agent activities
  • Slack/Teams Integration: Approve/reject directly from messenger

Competitive Analysis

CompetitorFeaturesWeakness
Peta DeskHuman approval workflowsEarly stage, limited features
In-houseFull customizationDevelopment cost, maintenance burden
NoneExecute without approvalSecurity/compliance risks

Opportunity: Absence of dedicated gateway solution based on MCP standard

Competition Intensity: LOW - Blue ocean market, first-mover opportunity

MVP Development

Timeline: 6-8 weeks

Tech Stack:

  • Language: TypeScript/Node.js
  • Framework: Fastify (high-performance proxy)
  • MCP Integration: @anthropic/mcp-sdk
  • Auth: OAuth 2.0 + RBAC
  • Notifications: Slack/Teams Webhook

MVP Features:

  1. MCP proxy server (intercept tool calls)
  2. Policy engine (YAML-based rule definition)
  3. Approval queue and web dashboard
  4. Slack notifications and quick approval
  5. Basic audit logging

Future Features:

  • ML-based anomaly detection
  • Auto-approval learning
  • Multi-agent session management
  • SOC 2 / GDPR compliance reports

Revenue Model

Model: Subscription (B2B SaaS)

Pricing Structure:

  • Starter ($49/mo): 5 agents, 1,000 requests/month, email notifications
  • Pro ($149/mo): 25 agents, 10,000 requests/month, Slack integration, team approvals
  • Enterprise (Custom): Unlimited, SSO, audit reports, SLA

Revenue Projections:

  • 6 months: $5K-15K MRR (with early adopter traction)
  • 12 months: $20K-50K MRR (with enterprise contracts)

Risk Analysis

RiskLevelMitigation
TechnicalLOWMCP standard stabilized, well documented
MarketMEDIUMDemand varies with agent adoption speed
ExecutionMEDIUMSecurity product requires reliability focus

Key Risks: Major platforms (Anthropic, OpenAI) might provide this feature natively

Who Should Build This

  • Developers with security/DevOps background
  • Those with AI agent project experience who felt the need for control
  • Experience in B2B SaaS sales and customer success
  • Familiar with proxy/gateway architecture
  • Domain knowledge in compliance and regulations

If you’re building this idea or have thoughts to share, drop a comment below!