Back
MCP Workspace Automation Agent

MCP Workspace Automation Agent

MCPOpenAIStreamlitAI Agents

I built an MCP-powered workspace agent that connected three productivity platforms—Jira, Slack, and Notion—through the Model Context Protocol. Users could give plain English commands like "Create a Jira ticket for the login bug and notify the dev channel" and the agent autonomously executed actions across all platforms.

Implemented an MCPManager class that handled stdio transport connections to multiple MCP servers, discovering available tools and resources via JSON-RPC protocol. Built an autonomous agentic loop using OpenAI GPT-4o that chained tool calls until task completion, with intelligent error handling and retry logic for failed operations.

Created a Streamlit interface with real-time tool execution logging, showing each step the agent took across platforms. This project taught me the architecture of MCP servers, cross-platform automation challenges, and the complexity of building reliable autonomous agents that coordinate multiple external services.

Background

Vinit skipped presentations and built real AI products.

Vinit Gujarathi was part of the August 2025 cohort at Curious PM, alongside 15 other talented participants.