Rows.page

What the rows.page MCP server does

The rows.page MCP server gives agents a way to publish tabular or structured data to rows.page for human inspection. A push can create a dataset or add a new version to an existing one. The response includes a URL that a person can open to view the data, filter rows, chart columns, run SQL, inspect pinned findings, and review changes between versions.

The service accepts CSV, TSV, JSON, JSONL, row arrays, and Parquet imports. Data can be supplied inline or fetched from a URL. Inline MCP payloads are limited to 16 MB; larger inputs should use the URL option. Viewing does not require signup, while an API key connects pushes to an account and enables account-level dataset management.

How it works

The rows.page MCP server is a remote, stateless JSON-RPC service at https://rows.page/mcp. It uses streamable HTTP, so there is no local package or process to install for clients that support remote MCP connections.

Each push is treated as an immutable version. Re-sending identical content reports that it is unchanged rather than creating another version. For meaningful row comparisons, provide an id_column; later pushes can then report added, removed, and changed rows, along with a URL showing the differences. A stable dataset key can be used with an API key for recurring updates.

Finding queries can use the rows.page query syntax for text, fields, null checks, prefixes, numeric ranges, dates, nested JSON paths, list tags, Boolean expressions, sorting, and selected columns. Queries beginning with SELECT, WITH, or FROM run as SQL against the data table.

Setup and configuration

Configure an MCP client with the endpoint URL. The documented clients include Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, Codex CLI, and Gemini CLI. Claude Desktop uses mcp-remote with Node.js; the other examples connect to the remote endpoint using their HTTP MCP settings.

Authentication is optional for anonymous pushes. To associate datasets with an account, create an API key at rows.page/account and send it as a Bearer authorization header or as the apiKey URL parameter. Account authentication is required by list_datasets and permits re-pushing by a stable key.

Tools and capabilities

The rows.page MCP server exposes these tools:

• push publishes data, creates versions, accepts optional titles and metadata, and can attach findings and provenance.
• pin_finding saves a titled view on a dataset the caller owns, including its filter or SQL, sorting, selected columns, and note.
• get_dataset returns metadata, recent versions, findings, columns, and the first 20 preview rows.
• list_datasets lists account datasets from newest to oldest and requires an API key.
• delete_dataset removes an owned dataset and immediately disables its URL.
• get_limits reports the plan, usage, and limits for file size, dataset count, and version history.

Limitations and notes

Anonymous datasets are deleted after seven days unless they are kept. Anonymous usage is limited to 50 pushes and 1 GB per day per IP, with 60 MCP calls per minute. Anonymous files can be up to 100 MB, although inline MCP pushes are capped at 16 MB. Paid and free plan limits may differ.

The project is MIT licensed. Its category is Data Visualization, but the service primarily transfers, versions, queries, and presents datasets; the supplied material does not describe a local visualization runtime or a separate chart-generation tool.