New video on the flespi YouTube channel! We recently introduced AI integration tools, and I spent a great time working with them – the result is a 30-minute demo video. It's a completely new way of working with flespi.
For me, the shift is this: I no longer need to hunt for the right API call, look up its schema, figure out which device type ID corresponds to my tracker model, or decide which plugin type fits the job.
I don’t spend time re-reading the docs or digging into the flespi.io interface (even though it's incredible!). What I do have is a business goal – a customer to onboard, a requirement to meet – and that’s where my attention belongs.
Translating that goal into a concrete flespi configuration is exactly where modern agentic tools like Claude Code, equipped with the flespi skill and MCP server, shine. The agent figures out the entities, picks the right endpoints, validates the schemas, and executes the calls. I approve the steps and watch it happen in the flespi panel in real time.
In this video, I play a project manager at a Telematics Service Provider company with a new customer to onboard. I give the agent a short project brief and let it drive – from architecture to a working setup to a reusable provisioning script – while I stay on the business layer the whole time (almost). You’ll see how:
- I equip my AI agent with the flespi skill and configure the flespi MCP server.
- A failure case: the consult-flespi-account tool call runs into a timeout; I grab the full response from /ai/logs, paste it back, and the agent continues from there.
- The agent proposes a best-practice architecture, creates every flespi entity needed (channel, subaccount, groups, plugin, stream), and validates the pipeline end-to-end on one test device with a synthetic message.
- The agent writes an automated device creation script and, noticing I have no token in the environment, proposes minting a separate one.
- At one point, the agent tries to break free (not really) – reaching for curl plus a Python one-liner on the raw OpenAPI spec instead of the flespi MCP schema tools – but I catch it and steer it back. :)
What I feel throughout this session – and what I want to highlight – is that an agentic assistant is much more than a chatbot or a remote control for creating entities: it wasn’t just executing my instructions, it was deciding on its own what questions to ask and which tools to consult along the way. You don’t hand it a list of steps – you hand it a project, and it becomes the expert who figures out the steps for you.