DATIUSDocs

AI assistants (MCP)

The Datius MCP server lets an AI assistant that supports the Model Context Protocol (MCP), an open standard, read your signals: list them, query them over time, compare periods, look at recent events and explain what changed. It's read-only, and the assistant you already use does the thinking. Included on some plans; the pricing page shows which.

Many assistants support MCP, among them Claude, ChatGPT, Cursor, VS Code with GitHub Copilot, Windsurf and Gemini CLI. There are two ways to connect:

  • By URL: add Datius as a connector and sign in with your Datius account. Nothing to install, and one connection covers every workspace you can see.
  • On your machine: run @datius-io/mcp (on npm) with a read API key for one workspace, for assistants that start MCP servers locally.

Connect by URL

  1. In your assistant, add a custom connector (or remote MCP server) with this URL, your Datius address followed by /api/v1/mcp (on datius.io, https://datius.io/api/v1/mcp):
Text
https://datius.io/api/v1/mcp
  1. When it asks, sign in to Datius and allow access. The assistant reads as you, with your permissions, in every workspace you're a member of.
  2. If you're in more than one workspace, name the one you mean (“in Meridian Market, …”); the assistant can list them with list_workspaces.

In Claude Code:

Shell
claude mcp add --transport http datius https://datius.io/api/v1/mcp

Other assistants that connect to remote MCP servers with sign-in take the same URL in their connector or MCP settings. If yours can't sign in, run the server on your machine instead (below).

Connected by URL, the assistant can also read your alert rules (list_alerts) and search these docs (search_docs). Remove the connector in your assistant to disconnect.

Run it on your machine

  1. In Datius, open API Keys & Tokens and create a key with Read data (the signals:read scope) in the workspace you want the assistant to read, and copy it. The key decides the workspace, so it's the only thing to set.
  2. Add the server to your assistant's MCP configuration. Find your assistant below.

Claude Desktop, Cursor, Windsurf and Gemini CLI

These assistants share the same configuration, each in its own settings file:

AssistantSettings file
Claude Desktopclaude_desktop_config.json (Settings > Developer > Edit Config)
Cursor.cursor/mcp.json in your project, or ~/.cursor/mcp.json for every project
Windsurf~/.codeium/windsurf/mcp_config.json
Gemini CLI~/.gemini/settings.json

Add this to that file:

JSON
{
"mcpServers": {
"datius": {
"command": "npx",
"args": ["-y", "@datius-io/mcp"],
"env": {
"DATIUS_API_KEY": "<your-read-key>"
}
}
}
}

VS Code with GitHub Copilot

Copilot in agent mode reads .vscode/mcp.json in your project. It names the block servers and adds a type:

JSON
{
"servers": {
"datius": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@datius-io/mcp"],
"env": {
"DATIUS_API_KEY": "<your-read-key>"
}
}
}
}

Claude Code

One command adds it:

Shell
claude mcp add datius \
-e DATIUS_API_KEY=<your-read-key> \
-- npx -y @datius-io/mcp

Other assistants

Any other assistant that starts MCP servers locally works the same way: it runs npx -y @datius-io/mcp with DATIUS_API_KEY set to your key.

Whichever assistant you use, if the key or plan isn't right, the server stops at startup and says why, so check your assistant's MCP log if the tools don't appear.

What the assistant can do

  • list_signals: your signals with their types, units, whether higher is better or worse, and their tags with descriptions. Filling in descriptions gives the assistant your words for things. Also your derived signals, with their formulas, which query_signal and compare_periods read like signals. And the workspace's business context, when its owners have written one.
  • list_tag_values: the values a tag has had recently, such as your store IDs, so the assistant can filter by the exact one.
  • query_signal: a signal over a time range, by hour, day, week or month, optionally grouped by a tag and filtered by tag values.
  • explain_change: why a signal moved compared with its normal level, ranked by tag value. The same numbers as Analysis in the app.
  • compare_periods: two time ranges side by side, overall or per tag value.
  • recent_events: individual events, for examples behind a number.

Ask things like “Why were refunds high yesterday?” or “Compare online and in-store revenue this week with last week.”

Your data

The server only reads: connected by URL, with your own permissions; on your machine, with the permissions of the key you give it, only in that key's workspace. Results go to the assistant you connect it to, so they're handled under that assistant's terms. Remove the connector, or revoke the key in Datius, to cut off access at any time.