Providers, context engineering, MCP extensions and recipes — a full end-to-end run of the open-source Goose agent.
What we covered
Agents stop being demos when they can touch your tools. We ran Goose — the open-source coding agent — through its full loop on real tasks.
- Providers and models — swapping backends, what changes in behavior, and how to pick one for offline vs. frontier work.
- Context engineering — what the agent sees, how it remembers, and where runs go off the rails.
- MCP extensions — giving the agent new tools via the Model Context Protocol, live.
- Recipes — packaging a working session into a repeatable workflow others can run.
What we built
A Goose session that went from blank machine to a verified change — tools wired, extensions loaded, recipe saved — with 142 learners following along on their own laptops.
Takeaways
- The agent loop is simple; the leverage is all in tools and context.
- MCP turns “my agent can’t do that” into a fifteen-minute extension.
- Recipes are the missing artifact: record the run, share the run, replay the run.
Join the next one
We run sessions like this every Saturday in Hyderabad — free, hands-on, and open to everyone. RSVP on Meetup and bring a laptop with Python 3.10+.
