MCP-first
Connect your agent.
We’re MCP-first. When public access opens, connecting is one config block — and every tool is built to be called correctly from its description alone, with REST underneath for everything that isn’t an agent.
How it fits together
One server. A namespaced tool surface. Self-describing responses.
The tools share one service layer, so the MCP and REST surfaces return byte-identical payloads — nothing an agent can reach on one surface is missing from the other. Tools are namespaced by category (stocks_*, macro_*) so the surface grows across categories without collisions, and each carries the bounds, error codes, and an example call in its own description.
What connecting will look like
Illustrative only — the public endpoint isn’t open yet, and shapes may change. Join early access to get a real endpoint.
{ "mcpServers": { "agent-datasets": { "type": "http", "url": "https://mcp.agentdatasets.example/mcp" } }}Drop-in for MCP clients such as Claude Desktop, Claude Code, and Cursor. The url above is a placeholder — you’ll get your real endpoint when early access opens.
The tools an agent gets
Price and quote tools exist in the internal MVP but aren’t part of the public surface — they’re license-gated (see Data coverage).
Documentation
Full documentation is coming.
It will cover the response envelope, cursor-based pagination, the error taxonomy, and every tool’s parameters and bounds — so a model can decide whether and how to call a tool without leaving its context. Want early access to the docs and the endpoint?