Canopy

Canopy Documentation

Canopy is the financial command center for agentic payments. As AI agents start moving money on your behalf — paying for APIs, settling invoices, buying compute, transacting with other agents — finance and engineering teams need a single place to set the rules, watch every action, and control the money. Canopy is that platform.

Bring your own agent. Canopy doesn't replace the agent you've built — it wraps it. Whatever framework you use today (Claude Agent SDK, an MCP host, Node.js, Python, Vercel AI, LangGraph, OpenAI Agents, the Anthropic API, or your own stack), you connect it to Canopy in a few lines and gain a policy + treasury + visibility layer around it.

For each agent you register, you set a spend cap, a recipient allowlist, and an approval threshold. For your organization, you fund one USDC treasury, see every transaction in real time, and approve the ones that warrant a human. Your agent calls canopy.pay() from the code you already have; Canopy evaluates the policy in real time, signs server-side, and never hands anyone a private key.

These docs are for the developers wiring an existing agent up to Canopy. If you're new, follow the three steps below in order.

The three steps

  1. Connect your agent in the Canopy dashboard. Give it a name, pick an existing policy or create a new one, and Canopy returns an agt_… ID for it.
  2. Wire up credentials. Either run npx @canopy-ai/sdk connect in your project root (browser-based consent → credentials + auto-detected MCP clients in one shot) or copy the org API key (ak_live_…) from Settings → API Keys and paste it with the agt_… ID into your .env.
  3. Paste a snippet into your existing agent code. Pick your framework on the Connect your agent page — Canopy supports plain Node.js and Python plus the major agent frameworks.

Once your agent makes its first request, it shows up as connected in the dashboard.

Where to go next