Labs · Adam Honoré

Agent Data Gateway

Licensed, metered market data for AI agents: who's entitled to what, what they used, what it would earn, and what the venues will audit.

An MCP server that checks every agent request against its license, meters usage, and turns a month of activity into three views a market data business runs on: usage by agent, revenue under three pricing models, and the per-venue report an exchange would audit.

How this was built

AI agents are becoming heavy consumers of market data, and data providers are starting to serve it through MCP servers. That brings back the questions a data business has always had to answer, now for software instead of people: who is licensed for what, what did each one use, how should it be priced, and what will the exchanges that supply the data audit?

This prototype answers them end to end. An MCP server, built with the official Python SDK, checks every call against an entitlement table (agent, venue, and real-time, delayed, or reference data) and logs it, whether it's allowed or denied. Three scripted agents generate a month of usage. The month is replayed under three pricing models and turned into the report each venue would audit: real-time versus delayed, display versus non-display, unlicensed attempts, and fees owed.

Everything is synthetic: fictional venues, companies, identifiers, and fees. It runs on a laptop with one command, and nothing in it depends on a particular cloud.

A prototype of market data licensing and metering for AI agents, October 2026. Runs on Scripted agents over MCP in Amazon Bedrock.

This demo is interactive and needs JavaScript.