Your AI agent can shop on Cubbie
AI agents are starting to make software purchasing decisions, and Cubbie is built for it. Any AI agent — Claude, ChatGPT, or your company's own procurement bot — can connect directly to Cubbie through the Model Context Protocol, the open standard adopted across major AI platforms, and search our catalog of roughly 30,000 business software products, compare options, analyze your existing stack and spend, and propose purchases.
The design is deliberately dual-use, not a replacement for human buyers: your agent does the legwork, but every agent-proposed order is priced by Cubbie's servers (never the agent), bounded by spending caps you set, and requires explicit human approval before any payment occurs. An agent cannot move money on Cubbie, period.
Connect in one line
Add Cubbie as an MCP server in any MCP-capable client and the catalog tools work immediately — no account needed:
https://www.cubbie.com/api/mcpAdd an API key as Authorization: Bearer cb_live_... (created in your Cubbie dashboard) and the agent can also work with your organization's own stack, spend, and purchasing.
How it works
The agent researches like a procurement team
One trusted catalog instead of 30,000 vendor websites: structured pricing, compliance certifications, comparisons, and peer evidence drawn from real buyer data with strict anonymity protections.
The agent proposes, with guardrails
Prices are computed server-side from the live offer — the agent never supplies an amount. Agent keys carry per-order and monthly caps plus category limits that you control.
A human always completes the purchase
Every proposal lands in your approvals inbox with the agent's reasoning. Approve it and pay through the normal checkout, or decline it. Every action is logged and every key is instantly revocable.
What agents can do
Anonymous — works the moment an agent connects:
Hybrid semantic + keyword search across the full catalog
Structured attributes, compliance, proof facts, and live offers
Side-by-side data for 2-4 products
With an organization API key:
The organization's tracked subscriptions, spend, and renewals
Monthly software-spend snapshots
Record a subscription, canonically linked to the catalog
Peer adoption and spend benchmarks, anonymity-floored
Propose a purchase — a human approves before any payment
Prefer REST
Everything the MCP server exposes is also available as a versioned REST API with a machine-readable contract at /api/v1/openapi.json, and /llms.txt summarizes every agent-facing interface. Full documentation, scopes, and examples live on the developers page.