Connect through MCP
Connect a compatible agent to operate Extralt and inspect your organization's ecommerce data. Use this MCP endpoint:
https://api.extralt.com/mcpMCP uses browser OAuth rather than an API key. During authorization, sign in to Extralt, select the organization the client should use, and approve access. The selected organization is taken from the OAuth grant; tool inputs never ask for an organization ID.
Connect a client
Extralt advertises both CIMD and dynamic client registration, so supported clients can register automatically. You do not need to create an OAuth client, enter a client secret, or configure scopes manually.
Codex
Add the server from a terminal:
codex mcp add extralt --url https://api.extralt.com/mcpIf Codex does not open the authorization flow during setup, run:
codex mcp login extraltSee the Codex MCP documentation for client settings.
ChatGPT
Enable developer mode, create an app from the Extralt MCP URL, choose OAuth with CIMD, scan the tools, then complete authorization. Availability and admin approval depend on your ChatGPT workspace. See the ChatGPT developer-mode guide.
Claude
Open Customize > Connectors, add a custom connector with the Extralt MCP URL, and leave the optional OAuth client ID and secret blank. Select Connect to authorize it. See the Claude connector guide.
Cursor
Add the server to the project .cursor/mcp.json or your global MCP
configuration, then authenticate it from Cursor's MCP settings:
{
"mcpServers": {
"extralt": {
"url": "https://api.extralt.com/mcp"
}
}
}See the Cursor MCP documentation for the current configuration location and controls.
To use a different organization later, reauthorize the Extralt connection in your MCP client and select the other organization on the consent screen.
Available capabilities
| Stage | What an agent can do |
|---|---|
| Extract | Prepare and inspect catalog sources; manage Runs and Schedules; start Imports; list, count, and export Captures |
| Enrich | Start, inspect, and stop Enrichments; list and inspect enriched Items |
| Explore | Inspect coverage, facets, current markets, Variants, Offers, and history; run the four packaged Analyses |
The client discovers exact tool names and input schemas from the MCP server. The tools use the same capability services and authorization rules as API v1, but are organized around agent tasks rather than REST routes.
Long-running work and exports
Catalog-source preparation, Extract Runs, Imports, and Enrichments return a durable ID immediately. The result tells the agent which status tool to poll. Runs and Enrichments can be stopped; source preparation and Imports continue until they complete or fail.
Capture exports return a short-lived download URL for JSONL or Parquet. The URL
expires after 15 minutes and can be passed directly to a download client such
as curl.
Access rules
An active organization can use read and write tools. During the read-only subscription grace period, clients can inspect and export data, stop active work, pause Schedules, and replay a completed idempotent request, but they cannot start new work. Suspended organizations cannot use MCP data tools.
MCP calls are direct operations. Review write and destructive tool calls in your client before allowing them.
The REST API remains available separately and uses organization API keys. Its OpenAPI document describes the REST contract, not MCP tool schemas.