> For the complete documentation index, see [llms.txt](https://docs.bipeye.com/product-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.bipeye.com/product-docs/overview/openapi/connect-bi.p.eye-to-claude.md).

# Connect BI.P.EYE to Claude

BI.P.EYE is also available as an MCP server, so Claude can explore and query your data directly.

* **URL:** `https://mcp.bipeye.com/mcp`
* **Authentication:** header `Authorization: Bearer <your organization API key>`. It is the same key used by this API; `X-API-Key: <key>` also works.

In Claude, open **Customize → Connectors → Add custom connector**, enter the URL, and add the authentication header.

#### Tools

All tools are read-only. Claude picks them on its own; you just ask questions in plain language.

| Tool               | Input                                                            | What it does                                                                                                                                                                                                                    |
| ------------------ | ---------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `list_models`      | none                                                             | Lists the data models in your organization, with their engine, schema, dataset count and relationship count.                                                                                                                    |
| `describe_model`   | `modelId`, optional `datasetIds`, optional `includeSampleValues` | Returns the datasets in a model, their columns and types, sample values for low-cardinality text columns, and the relationships between datasets. Use `datasetIds` for large models or to get the rest of a truncated response. |
| `query_model`      | `modelId`, `question`, optional `previousQueryId`                | Answers a question about the model's data using the BI.P.EYE chat engine. Returns a `queryId`, a short answer, the decisions and assumptions made, the result rows and a download link.                                         |
| `get_query_result` | `queryId`                                                        | Returns the result of a query that was still running when `query_model` returned.                                                                                                                                               |

#### Results

* Results with up to 200 rows are returned in full. Larger results return the first 20 rows plus a time-limited link to download the full table.
* If a query takes longer than about 50 seconds, `query_model` returns `status: "running"` and Claude fetches the result with `get_query_result`.
* Every answer lists the decisions and assumptions behind it, such as which dataset and date column were used, so you can check how a number was calculated.

#### Follow-up questions

Ask follow-ups the way you would in a chat, for example "show me only the last 3 months" or "add the customer name". Claude sends the earlier `queryId` as `previousQueryId` with only the change, and BI.P.EYE refines that result instead of starting over. A follow-up must use the same model as the result it changes.

#### Access and data

* The organization comes from the API key. Claude can query every model in that organization, so create a dedicated key for Claude and delete it when no longer needed.
* Query results returned to Claude become part of the Claude conversation.
* BI.P.EYE keeps the question, the generated query configuration and an audit record of each request. Result rows are stored only in the temporary download file.

The MCP endpoint is not a REST API, so it is not listed below. MCP clients such as Claude discover its tools automatically.
