Overview

Aesops runs an MCP (Model Context Protocol) server that exposes the dataset catalog as tools any MCP-compatible AI client can call directly. Ask a client to find a dataset, inspect its schema, or run a query — the same catalog behind the REST API and Python SDK, reachable from inside a conversation.

Endpoint

https://aesops.co.ke/api/mcp

Streamable HTTP transport — add this URL to any MCP client that supports it directly. For stdio-only clients, use mcp-remote as a bridge.

Connecting a client

Just add the endpoint URL — no auth needed to connect or to browse:

{
  "aesops": {
    "url": "https://aesops.co.ke/api/mcp"
  }
}

Querying a dataset's actual data is gated (see below). When it is, if the client supports OAuth discovery (Claude Desktop, Claude Code) it detects the requirement automatically and opens a browser to sign in — nothing extra to configure. See Authentication for the full picture, including the API-key alternative for scripted or headless clients:

{
  "aesops": {
    "url": "https://aesops.co.ke/api/mcp",
    "headers": {
      "Authorization": "Bearer <your-key>"
    }
  }
}

Public vs. gated

Same split as the REST API:

• search_datasets, list_facets, get_dataset, and get_chart_colors are fully public — no auth is checked. The full catalog (keyless and gated datasets alike) is browsable for discovery.
• query_dataset_rows, aggregate_dataset, and distinct_dataset_values — the tools that return actual dataset rows — require an authenticated identity (OAuth session or read-scoped API key), unless the dataset is currently "keyless", in which case they work with no auth at all. An unauthenticated call against a non-keyless dataset gets back a plain tool error, not a connection-level failure — see Authentication.

See Tools for the full reference.