MCP server

Use Queria from MCP clients such as Claude Desktop and Cursor

queria mcp is a stdio MCP server. Any MCP-compatible client can explore and query Queria’s data through it.

For coding agents with shell access, such as Claude Code and Codex, we recommend the Queria skill (plugin) over MCP. This page is for MCP clients where the agent has no shell, such as Claude Desktop.

Setup

Add the configuration for your MCP client.

{
  "mcpServers": {
    "queria": {
      "command": "uvx",
      "args": ["--from", "queria[mcp]", "queria", "mcp"]
    }
  }
}

codex mcp add queria -- uvx --from 'queria[mcp]' queria mcp

Editing ~/.codex/config.toml directly works the same:

[mcp_servers.queria]
command = "uvx"
args = ["--from", "queria[mcp]", "queria", "mcp"]

claude mcp add queria -- uvx --from 'queria[mcp]' queria mcp

Provided tools

Tool	Purpose
`list_datasets`	List datasets
`search`	Keyword search across datasets, tables, and columns
`get_dataset_info`	Dataset metadata (license, source, etc.)
`get_schema`	List tables
`get_columns`	List columns
`query`	Run SQL
`list_docs`	List this documentation’s pages
`show_doc`	Fetch one documentation page as Markdown
`submit_feedback`	Report a dataset request, a data defect, or a documentation error
`set_feedback_consent`	Record whether agents may send feedback

For detailed arguments and behavior, see the MCP tools reference.

The MCP server uses the same token resolution as the CLI (the QUERIA_TOKEN environment variable → the config file). Save a token with queria login (browser approval) or queria auth set-token <token>, or pass the QUERIA_TOKEN environment variable in your MCP client’s configuration (see Authentication).

Result limits

query results are truncated by row count (default 100, max 1000) and payload size (about 1MB). This is by design, to protect the agent’s context. When truncated, the response carries truncated: true with a hint — narrow down with aggregation or LIMIT, or use the CLI’s --out for bulk extraction.

Agent-readable docs

From the CLI

queria docs list names every page, and queria docs show <name> prints one as Markdown.

queria docs list
queria docs show reference/mcp-tools

See the CLI reference for the options.

From the MCP tools

An MCP client reads the same pages through the list_docs and show_doc tools. See the MCP tool reference for their arguments.

Straight from a URL

The documentation site is also served in formats that need no HTML parsing.

https://docs.queria.io/llms.txt — index of pages
https://docs.queria.io/llms-full.txt — all pages concatenated
Append .md to any page path for raw Markdown (for example, https://docs.queria.io/reference/mcp-tools.md)

queria --help, queria --version, and the MCP server’s instructions all point at these ways of reading the docs.