Engagement summary:

Join a small engineering team building custom internal MCP (Model Context Protocol) servers in Python on Kubernetes, connecting enterprise systems such as Salesforce, JIRA, and Snowflake to internal AI/LLM applications. The shared framework, CI/CD, and platform infrastructure already exist; this role focuses on designing and implementing production-grade integration servers on top of them.

Responsibilities

- Design and implement MCP servers in Python using the MCP SDK (e.g., FastMCP) on the existing internal framework and scaffolding template.

- Author per-server specifications: define a focused v1 tool set (5–10 tools), identity and auth model, rate-limit/quota strategy, and pagination/truncation rules, and drive stakeholder sign-off before build

- Build robust async API clients (httpx or similar) for enterprise SaaS and data platforms, with retries, backoff, circuit breakers, and clean mapping of downstream errors (401/403/429, entitlement failures) into meaningful MCP responsesß

- Define tool interfaces with Pydantic schemas and model-friendly descriptions; iterate based on LLM evaluation runs to eliminate ambiguous or model-hostile tool designs

- Integrate with the organization's OAuth abstraction layer for per-user and service-account identity flows, and with secrets management for credential handling — never handling raw secrets in code or logs

- Implement caching, internal rate limiting, and quota-protection logic for metered vendor APIs

- Write unit and integration tests (including sandbox-tenant testing) and participate in LLM eval passes as part of the standard definition of done

- Support security reviews: least-privilege scope design, audit logging (user → tool → resource), guardrails on write/destructive actions, and prompt-injection-aware handling of untrusted data returned from downstream systems

- Deploy through the existing Kubernetes pipeline (Helm/Kustomize), configure dashboards and alerts (error rates, latency,



quota burn), and support two-week pilot rollouts per server

- Document each server in the internal catalog and contribute improvements back to the shared framework and scaffold template

Required qualifications

- 5+ years of professional backend development experience, with 3+ years in Python (async programming, typing, Pydantic)

- Strong track record integrating third-party REST APIs at production scale: authentication flows, pagination, rate limiting, webhooks, and error-handling patterns

- Working knowledge of OAuth 2.0 concepts (authorization code flow, token refresh, scopes) sufficient to consume an internal auth abstraction correctly

- Hands-on experience deploying and operating services on Kubernetes (containers, Helm or Kustomize, health probes, HPA basics) and working within established CI/CD pipelines

- Solid testing discipline: pytest, mocking external APIs, integration testing against sandbox environments

- Experience with observability tooling (structured logging, Prometheus metrics, OpenTelemetry tracing or equivalent)

- Strong written communication — this role produces specs and catalog documentation, not just code

Preferred qualifications

- Direct experience with MCP, LLM tool/function calling, or building AI agent integrations

- Prior integration work with one or more target systems (Salesforce APIs, Atlassian/JIRA, Snowflake, ServiceNow, Workday, or similar enterprise platforms)

- Familiarity with secrets management (Vault, External Secrets Operator) and enterprise security review processes

- Experience with Redis or similar for caching strategies against metered APIs

- Exposure to prompt-injection risks and secure design patterns for LLM-facing services

Success measures (first 90 days)

- 3–5 servers shipped to production through all quality gates with pilot sign-off

- Zero security-review findings related to secret handling or scope over-provisioning

- LLM eval pass rates meeting team standards without repeated malformed tool calls

📌 AI Engineer (México)
🏢 Entellux
📍 México