Build an agent that queries open data without paying for tokens
The architecture behind our Open Data Agent, built live.
Date to be announced · Save your seat
Save your seat →
- Cost
- Free
- Format
- Online and live
- Length
- 90 min
- Date
- Date to be announced
An agent is not a model that only answers: it is a model with tools that takes actions in a loop. The hard part is not the model call, it is designing the tools, their contracts and their limits well.
In this workshop we build, step by step, an MCP server that gives a Claude agent tools to query public data. The model runs on the side of whoever uses the agent, so your server never touches a single token. We cover contracts, validation, limits and what to do when something fails.
What you take away
- 01Well-designed toolsJSON Schema contracts, a single responsibility per tool and descriptions the model understands.
- 02Safe queriesRead-only access, row and time limits, and validation before any query runs.
- 03An architecture with clear costsHow to keep the model cost on the side of whoever uses the agent, with your server never touching tokens.
Who it is for
- Technical teams that want to build their first agent
- Developers who work with data and APIs
- Anyone who wants to understand MCP beyond a demo
- People who do not code
Bring ready
- A laptop with internet
- Basic Python: functions, dictionaries and error handling
- Some SQL helps, but is not required
Agenda
- 10 MINWhat an agent isThe difference between a model that answers and a model with tools that acts in a loop.
- 20 MINTools and contractsEach tool's name, description and input schema, and how the model chooses between several.
- 25 MINThe MCP server, liveWe build the tools that query the public data and connect them to the agent.
- 20 MINLimits and errorsValidation, read-only access, row limits, retries and timeouts so the agent does not break.
- 15 MINCost and next stepsWho pays for tokens in this architecture and what it takes to go further.
Questions
01How much does it cost?
Nothing. The workshop is free.
02How does it work?
It is online and live. Once the date is set, we email you the link.
03Can you run it inside my company?
Yes. Write to us and we can talk about adapting it to your team and its data.
Build your first agent with sound judgment
Sign up and we will email you the link once the date is set.
Save your seat →Want it just for your team?
We run any of these workshops inside your company, with your processes and sample data. It starts with a 45-minute assessment of where AI pays off, on us.