MCP/Integration Developer
Summary
Build and operate MCP servers and AI agents for enterprise consumption: deploy FastMCP- or Workato-based MCP servers to Google Cloud Run, implement per-user OAuth 2.0 authentication, and build ADK agents on Vertex AI Agent Engine registered in Gemini Enterprise. Core stack is Python, Google Cloud (Vertex AI, Cloud Run, IAM), and the MCP protocol.
This is remote role for India applicants.
Build MCP servers, either leveraging FastMCP built deployed to Cloudrun, or using a service like Workato to build them.
Own the full consumption chain: how an ADK agent (and/or Gemini Enterprise directly) discovers, authenticates to, and calls each server — JSON-RPC over Streamable HTTP, tool discovery, and the session handshake.
Implement per-user, permission-aware authentication so every call runs as the asking user and the upstream system enforces that user's own permissions — OAuth 2.0 authorization-code flow with PKCE, refresh tokens / offline access, and secure token forwarding.
Build and deploy agents on Vertex AI Agent Engine using ADK, wire their tools and OAuth authorizations, and register/publish them into Gemini Enterprise for testing MCP.
Operate what you ship: monitor and debug MCP servers and agents in production across the whole chain (auth handshake, tool discovery, agent runtime).
Follow secure-by-default and governance practices: least privilege, no secrets in code, private/network-restricted services, cataloging and access control.
REQUIRED SKILLS (MUST HAVE)
End-to-end MCP fluency — not only building a server, but how agents consume it: the protocol (JSON-RPC / Streamable HTTP), tool discovery, and the session handshake.
Hands-on OAuth 2.0 / OIDC for agent scenarios — authorization-code flow, PKCE, refresh tokens, scopes, and per-user token forwarding so calls run as the signed-in user.
Hands-on Google Vertex AI — Agent Engine, ADK, Gemini models — plus Cloud Run, IAM / service accounts, and Secret Manager.
Proven production-grade delivery (reliability, maintainability, scalability) — not just prototypes.
Enterprise security discipline — least privilege, no secrets in code, private services, no token leakage, fail-closed design.
Strong Python (or comparable) for AI and cloud development.
5+ years senior software development, with 2+ years building AI/agent solutions in an enterprise environment.
NICE TO HAVE
Registering agents / MCP servers into Gemini Enterprise and/or a central Agent Registry.
iPaaS platforms (Workato, MuleSoft, etc.) and how their generated connectors/MCPs fit a governed agent architecture.
Advanced prompt engineering for agent behavior.
Prior work on a Corporate/Enterprise AI team building internal tools