Skip to content

AI access

Let an assistant query the data directly

Boldtive exposes its dataset over the Model Context Protocol. An AI assistant can answer questions from the data itself, under the same permissions and limits as any other client, with every call logged.

Why this is different from pasting data into a chat

The usual way to get an assistant to reason about a dataset is to export it and paste it in. That makes a copy nobody controls, it is stale the moment it is made, and the assistant cannot tell which rows were reviewed.

A connection removes all three problems. The assistant queries the live dataset, sees only what your account permits, and every call is recorded against the connection that made it. Disconnecting it in your portal ends the access straight away.

The endpoint

One endpoint. In Claude or ChatGPT you add it as a custom connector and sign in with your Boldtive account, with no key to paste. Other clients can send an API key instead.

Endpoint
https://api.boldtive.com/mcp

The four tools

  • search_crop_availability

    Returns individual records, each recorded by a collector and approved by a reviewer, matching crop, variety, grade, quantity, area and date filters.

  • get_supply_summary

    Returns totals grouped by crop, ward or council. The assistant is steered toward this one first, because a summary answers most questions at a fraction of the cost.

  • list_reference_data

    Returns crops, varieties, grades, units and the administrative tree, so the assistant can build a valid query.

  • get_coverage

    Returns where data exists and how recent it is, so the assistant can tell an empty result from an uncollected area.

What a question looks like

The analyst asks in plain language. The assistant chooses a tool, queries, and answers from what came back.

Asked

"Where is there grade A tomato available in the next two weeks, and roughly how much?"

Answered from

get_supply_summary(crop: "Tomato", groupBy: "council")

The assistant returns the totals per council with the date window applied, and can follow up with search_crop_availability if the analyst wants the individual records behind a figure.

Illustrative exchange, not a transcript

What it works with today

Works in Claude and ChatGPT with one sign-in, and in any client that can send a header.

In Claude and ChatGPT, add the endpoint as a custom connector, sign in to your Boldtive account and approve the connection. It then appears in your portal, where you can disconnect it at any time. Custom connectors depend on your plan with Claude or ChatGPT, so check that yours allows them.

Clients such as Claude Code, or anything else that can send a header, can use an API key instead. We would rather say all of this here than have you find out after signing something.

The same limits apply

An assistant is a client like any other. It cannot see farmer names, phone numbers or exact coordinates, because those are never in a response. It is subject to the same rate limit as your other keys, and every call is logged.

Want to connect an assistant?

Tell us which client your team uses and what they need to ask. We will tell you whether it works today.

Request access