Use a dataset from your agent
Connect Claude Code, Codex or ChatGPT to Jsonify over MCP, then ask questions of live data and ask Jason for new work without leaving your tools. About five minutes.
Everything in your workspace is available to the assistant you already use. Once connected, your agent can list pipelines, read and query datasets, start runs, and hand a brief to Jason. This tutorial connects one client and runs three questions against the Hacker News dataset from the first tutorial.
- Server URL
- https://factory.jsonify.com/mcp
- Auth
- OAuth, sign in once
- Transport
- Streamable HTTP
- Works with
- ChatGPT, Claude, Codex, Cursor, Copilot
1. Connect
Pick your client. The full per-client steps are on Connect your agent; the short version:
Sign in with the same account you use in Jsonify. The connection is scoped to your organisation; you choose the workspace in conversation.
2. Ask a question
In my Jsonify workspace “Latest Hacker News Posts”, which domains were linked most often, and what was the highest-scoring post from each?
Your agent lists the workspace, finds the dataset, runs a query, and answers with a short table. It cites the dataset and the version it read.
The agent reads the active dataset version. Ask for “the version from Monday” and it will read that one instead.
3. Start a run and wait for it
Run the Hacker News pipeline now and tell me when it finishes.
The agent starts a run and polls it until it completes, then reports rows collected and links the run. Runs started this way are ordinary runs: they appear in the runs rail, publish a new version and count rows like any other.
4. Ask Jason for something new
Ask Jason to add a “domain” column to the Hacker News dataset, extracted from the url, and to keep the existing rows.
Your agent hands the brief to Jason and returns a link to the conversation. Jason makes the change as a new version and verifies it, exactly as if you had typed the request in Jsonify. If Jason has a question, your agent relays it and you answer in place.
Your agent cannot approve things on your behalf. Anything that needs a decision, such as accepting a proposal or confirming a credential was set up, comes back to you as a question.
What the agent can do
| Ask for | What happens |
|---|---|
| Workspaces, pipelines, datasets, runs | Listed with names, ids and status |
| A dataset’s rows or a query | Rows are returned in pages; large results are cut off with a cursor |
| A pipeline run | Starts a run; can wait for it |
| New work or a change | Opens a Jason conversation and returns its link |
| A conversation’s progress | Reads Jason’s latest events and questions |
| Cancel | Stops the running Jason turn |
The full tool list is on Connect your agent. For scripts and automation rather than conversation, use the command line.