Skip to main content

Quickstart

You’ll need the medatarun command installed. See the installation guide for setup.

Create a directory that will hold your project data, then start the server ▶️ medatarun serve

Resources and commands​

Medatarun exposes a set of resources. Each resource provides commands, sometimes with parameters.

UI, CLI, REST API or an AI agent using the MCP server all rely on the same model.

🌐 Use with UI: human understanding and manual management​

UI provide business and data users a clear view of your models. Use it in your meetings, reviews as a based and shared understanding.

Open http://localhost:8080 to view and manage your models.

The command palette lets you run resource commands and provide parameters when needed.

For example, we can create a model by importing an existing schema. Open the Commands palette, choose model > import and in the "from" field add this URL. This is a TableSchema-style model definition (in French, everything in French and complicated).

On the Models page, you'll see a new available model, clic and browse entities, types and documentation. This will help you explain the data in this CSV file.

Everything is French and you don't understand anything? See below.

⭐ Use with your AI agent: the fun and productive way​

Optional: update your AI AGENTS.md​

If your AI Agent doesn't support MCP Instructions (see column Instructions on this table) you'll need to create or add usage instructions in an AGENTS.md file. Otherwise, you AI may act strange.

echo "# AGENTS.md" > AGENTS.md
curl -s http://localhost:8080/api/config/aiAgentsInstructions >> AGENTS.md

Use with your agent​

Run your AI Agent; it will call Medatarun through MCP on its own. For example:

Imagine a business user starting something:

Can you set up a ‘contacts’ thing? I just want people with their email, name, job, phone if we have it, LinkedIn, 
and which company they work at. For each company, keep its name, website, domain, and a short description of what it does.

Note how "not very precise" the request is, at least, no as precise as a developer or data analyst would.

Extend the model with company employee counts and Glassdoor-like information.
I am a business user, explain "contacts";
I am a developer, explain "contacts";

Another user, a developer, can continue work to adjust the model, entities, types

Make linkedin_url required. Replace the ShortDescription type with a new Markdown type

And then

Enrich company description. Explain that we need information of main activities and how the company relates to us (customer/partner/vendor/prospect/competitor).

Remember you can view and browse everything using user interface.

Also, you remember the French data example? what if you just ask:

Translate model "datapackage-irve", entities and attribute name and descriptions (do not change ids)?

As everything is version controlled, if AI goes crazy, you can always rollback and track changes on GIT.

🌐 Use with RestAPI​

  • List available resources, commands and arguments curl http://localhost:8080/api | jq
  • Execute a command on a resource by POST on http://localhost:8080/api/<resource>/<command> and pass parameters in Json format in the body.
  • Rest API : curl http://localhost:8080/api/model/createModel?id=mymode, it's the same commands as CLI, you can use GET and send query parameters or POST with a form body.
  • Rest API : curl http://localhost:8080/api/model/inspect

🖥️ Use with CLI : CI/CD or manual mode,​

  • medatarun help: get basic help and available resources
  • medatarun help <resource>: list available commands on a resource, for example: medatarun help model
  • medatarun help <resource> <command>: get description of a command, for example: medatarun help model inspect
  • medatarun <resource> <command>: runs a command, parameters are passed by --key=value. Complex values shall be in Json format.