AI integration · Featured case study · 2026

Secure AI Integration Gateway for HRM & CRM

A production Model Context Protocol service that gives AI clients useful access to business workflows while preserving authentication, validation, and operational safety.

role
Primary engineer
period
2026
stack
TypeScriptNode.jsMCPZodSOAPHTTPIIS
// 01 context

The challenge

The gateway had to bridge AI clients with established HRM and CRM interfaces without widening the trust surface. Tool inputs, remote responses, credentials, sessions, attachments, and state-changing operations all needed explicit boundaries.

// 02 delivery

What I contributed

  1. 01

    Designed the TypeScript service from HTTP hosting through MCP tool registration, application services, integration gateways, and external clients.

  2. 02

    Exposed focused capabilities for jobs, candidates, attachments, companies, users, tasks, work orders, custodians, and controlled status changes.

  3. 03

    Introduced schema validation for environment configuration, tool input, and every external response boundary.

  4. 04

    Built safe write workflows with pre-write reads, stale-state checks, post-write verification, and sanitized error reporting.

  5. 05

    Prepared IIS hosting, TLS, rollout, integration, and implementer documentation alongside broad automated coverage.

System viewSecure AI Integration Gateway for HRM & CRM
AI CLIENTSMCP / HTTPGATEWAYAUTH · LIMITSVALIDATIONSERVICESREAD · WRITEVERIFYBUSINESSHRM · CRMSCHEMA BOUNDARYAPPLICATION BOUNDARYSANITIZED OPERATIONS
// 03 architecture

Key decisions

01

Validate where trust changes

Configuration, MCP arguments, SOAP payloads, and webhook responses are parsed at their owning boundary before entering application logic.

02

Verify every write

State-changing tools read current state, apply one explicit update, then read again to confirm the saved value.

03

Keep composition explicit

Small services depend on narrow gateway contracts and are assembled in one composition root without reflection-heavy infrastructure.

04

Treat operations as product work

Request limits, lifecycle handling, deployment guidance, sanitized diagnostics, and integration contracts ship with the application.

// 04 result

The outcome

The result is a narrow, testable AI integration surface that can support useful HRM and CRM operations while keeping failure behavior, write safety, and production ownership visible.

// contact

Have a complex system? Let’s make the next change safer and more useful.