Hexabot + OpenCode: build an agent with MCP and Gemini

This tutorial builds a study abroad assistant with Hexabot, OpenCode and the Gemini API. The speaker explains an AI agent through intent detection and tool selection: a question triggers a country-data or university lookup, then Gemini turns the API response into an answer.

The setup starts with the Hexabot CLI, a project folder and a local server at localhost:3000. After a basic greeting workflow, the speaker installs Hexabot Action Creator and Hexabot Workflow Writer skills. These give the coding assistant guidance for generating actions and workflows. OpenCode runs inside the project folder and connects to Hexabot through MCP using a token and a configuration file.

The walkthrough includes connection failures and credential troubleshooting. A Gemini API key goes into Hexabot's credentials, while the model step references the credential record ID rather than the key itself. The speaker then tests country information and university searches, using execution logs to inspect which tools ran and what they returned.

This is a self-hosted workflow setup with external model and data APIs; the demonstrated assistant depends on those services. The speaker warns that model requests can encounter rate limits, which appear in the execution logs. The example demonstrates workflow construction and testing rather than establishing production readiness.