Show HN: Built an open-source SDK to simplify tool authentication for AI Agents (github.com)

Agentor is a new open-source SDK and framework that makes it easy to wire LLM-based agents into real-world tools like Gmail, Google Calendar, CRMs and more. Installable via pip, it offers a CLI, web UI, and a Python API for both conversational workflows and direct tool calls (e.g., search Gmail, list calendar events). Pre-built agents and an OpenAI App SDK speed development, while features like desktop OAuth, automatic token refresh, per-user credential isolation, and local secure storage aim to keep user data private and auditable. Its key technical innovation is AgentMCP, a unified tool-routing layer that presents multiple underlying services to the model as a single tool. That prevents context bloat and degraded LLM performance from stuffing many tool APIs into prompt context—AgentMCP routes requests to the correct integration based on intent. You can also bypass LLMs entirely and call integrations directly (e.g., GmailService.search_messages), and the API supports model overrides, multi-turn agents, and programmatic OAuth flows for public applications. For builders, Agentor is pragmatic: it supports Gmail and Calendar today, with email-sending, Slack, HubSpot and Document AI integrations planned; it's built on OpenAI Agents and Google APIs and released under Apache 2.0. The combination of secure local credential management, routing to reduce LLM load, and easy developer ergonomics makes it a useful foundation for production agent deployments that need to access personal or enterprise data.