Connect your agent (MCP)

Jsonify is an MCP server. Connect it to the assistant you already use and that assistant can read your datasets, inspect pipelines and runs, start a run, and hand Jason a brief, all through the same account and the same workspaces you see in the browser.

Server URL

https://factory.jsonify.com/mcp

Transport

Streamable HTTP

Sign-in

OAuth, with your Jsonify account

Scope

The organisations you authorise

What you can do from an assistant

• Read data. List datasets, look at a dataset’s columns and versions, and pull rows with a filter. The assistant can then reason about them, chart them, or fold them into whatever it is doing.
• Inspect pipelines and runs. See what a pipeline does, its active version and schedule, and how a run went: status, duration, and what it published.
• Start a run. Run an activated pipeline with a row cap.
• Ask Jason for work. Build a new pipeline, change an existing one, or ask “why did last night’s run fail?” Jason works in a normal Jsonify chat; the assistant relays questions and the outcome.
• Ask questions. “Which retailer raised prices most this week?” becomes a dataset read and an answer, in your assistant.

What it cannot do: run SQL, upload files, change schedules or settings, or edit the pipeline itself. Those stay in Jsonify, or go through Jason as a request.

Connect your client

Every client needs the same two things: the server URL, and a sign-in with your Jsonify account. Pick your client.

ChatGPT

1. Turn on Developer mode. Open Settings → Security and login and enable Developer mode.
2. Create the connection. Open Plugins and select +. Name it Jsonify, add a short description, and paste https://factory.jsonify.com/mcp.
3. Use OAuth. Choose OAuth for authentication, select Create, and sign in to your Jsonify account when asked.
4. Enable it in a chat. Start a new chat, select Jsonify from + → More, and describe the dataset you want.

If Developer mode is unavailable, your plan or workspace settings may restrict custom connections.

Claude

Claude.ai and the desktop app:

1. Open connectors. Go to Settings → Connectors.
2. Add a custom connector. Select + → Add custom connector. Name it Jsonify and paste https://factory.jsonify.com/mcp.
3. Connect. Select Add, then Connect, and sign in to Jsonify when asked.
4. Enable it in a chat. In a new chat, open + → Connectors and enable Jsonify. Then describe the data you need.

On a Team or Enterprise plan, your workspace owner may need to add the connector first.

Claude Code, in a terminal:

claude mcp add --transport http jsonify https://factory.jsonify.com/mcp
claude mcp login jsonify

The second command opens a browser to sign in. Choose the organisation the assistant should have access to.

Copilot and Teams

Microsoft 365 Copilot reaches Jsonify through an agent built in Copilot Studio.

1. Add the tool. Open your agent in Copilot Studio. Go to Tools → Add a tool → New tool → Model Context Protocol.
2. Point it at Jsonify. Name it Jsonify, paste https://factory.jsonify.com/mcp, and configure OAuth 2.0. Create a connection, sign in, and add the tool to your agent.
3. Publish. Test the agent, then publish it. Open Channels → Teams and Microsoft 365 Copilot, enable Make agent available in Microsoft 365 Copilot, then select Add channel.
4. Install in Teams. Choose See agent in Teams → Add. In Microsoft 365 Copilot, type @, select your agent and ask your question.

Requires Copilot Studio access. Your administrator may need to configure OAuth or approve the agent.

Codex

codex mcp add jsonify --url https://factory.jsonify.com/mcp
codex mcp login jsonify

Cursor

In Cursor’s MCP settings, add a new server named Jsonify of type Streamable HTTP with the URL https://factory.jsonify.com/mcp, then complete the sign-in when Cursor opens the browser. The equivalent configuration entry is:

{
  "mcpServers": {
    "jsonify": { "url": "https://factory.jsonify.com/mcp" }
  }
}

Other assistants

In your assistant’s connections settings, add a remote MCP server named Jsonify. Use Streamable HTTP and OAuth, then sign in to Jsonify. The assistant must support remote MCP servers with OAuth; API-key-only clients cannot connect.

The Add integration catalogue in Jsonify has an Assistant connection card that shows these same commands for Claude Code, Codex or another assistant, and a Check connection button that confirms the sign-in worked. It is the quickest way to verify a new connection.

The organisation requirement

An assistant connection is a grant on your account, made through your Jsonify sign-in, for one or more organisations. Three consequences:

• You need a Jsonify account with an organisation. Sign up in the browser first if you have not; the connection then sees every workspace in the organisations you authorise, subject to your membership at the time of each request. Leave an organisation and the assistant loses it too.
• Brand-new users can start from the assistant. If you sign in during the connection without an organisation yet, the assistant can still describe what you want. Jason prepares a brief, and the assistant gives you a link to review it, accept the terms, create your organisation and start the first build in the browser. Reconnect afterwards so the assistant is authorised for the new organisation.

Starting from an assistant without an account

If you connect before you have a Jsonify organisation, the flow looks like this:

1. Describe what you want in the assistant, as you would to Jason. The assistant starts a preparation and relays Jason’s questions.
2. Open the link the assistant gives you. It shows the brief Jason prepared and the free allowance, asks you to accept the terms, creates your organisation, and starts the first build in the browser.
3. Reconnect the assistant. Sign in through the connection again so it is authorised for the new organisation. The assistant picks up the same workspace and chat rather than starting another build.
4. Review and accept. When the first run has published, look at the dataset. Telling the assistant you are happy marks the workspace live, the same as Open workspace in the browser.

A preparation waits up to 24 hours for you to open the link; after that, start again.

What the assistant can call

Assistants choose tools on their own from your request; you do not name them. This table is for knowing what is possible and for reading the assistant’s activity log.

Reads are bounded so a chat never drowns: fifty items per page for lists, one hundred rows and fifty columns per dataset read, twenty filters, and a pinned dataset version so paging is consistent. The assistant pages when it needs more.

run_pipeline starts existing, activated pipelines only, always with a row cap; it does not change parameters or schedules. For anything that changes the pipeline, the assistant uses request_work, and the change goes through Jason’s normal build, review and activation.

Examples

Show me the Live product offers dataset in Market Radar. Which five products had the largest price change since the previous version, and who sells them?

The assistant reads the dataset, pulls the rows it needs, and answers in the chat. Nothing changes in Jsonify.

Run the price promotion tracking pipeline, capped at 200 rows, and tell me when it finishes and how many rows it published.

The assistant starts the run and reads it back when it completes. The run appears on the pipeline’s runs list like any other.

Ask Jason to add a “promotion ends” column to the price promotion tracking pipeline, using the end date shown on each retailer’s offer badge. Let me know if it asks anything.

The assistant sends the brief to Jason and gives you the chat link. If Jason has a question, the assistant relays it and you answer in the same conversation. When the change is reviewed and activated, the assistant tells you the new version.

Chats started from an assistant

Every request an assistant makes through request_work is a real chat in the workspace. It appears in the chat history with the assistant’s name and icon as the source, so a teammate looking at the history can see what was asked from where. You can open it in the browser at any time, add to it, or stop it. Closing the assistant does not stop work already handed to Jason.

Managing connections

Assistant connections are listed at the bottom of the Integrations page, in their own section and under the Assistant connections category filter. Each card shows the assistant’s name, “Connected” or “Access blocked”, and a menu with Connection details and Disconnect.

Connection details shows the facts: the owner (your account), the access (the organisations you authorise, subject to your current membership; not limited to the workspace you have open), the server URL and the client identifier.

Disconnect removes the assistant’s authorisation for your account across all organisations. Tokens it already holds stop working straight away. Chats it started and work Jason already accepted remain; stop active work from its chat if you need to.

• Connections belong to you, not the workspace. They cannot be moved into folders or bulk-deleted with workspace integrations; open the card to disconnect.
• “Access blocked” means Jsonify has cut the connection off but the assistant has not yet released its grant. Press Disconnect again to finish.

An assistant connection carries your whole membership. If you share a computer with someone, disconnect before you hand it over.

Troubleshooting

What’s next