Capability-Based Authorization for AI Agents
Identity tells you who is calling. Capability auth tells you what this agent can access, spend, delegate, and revoke before every API or MCP request.
Why identity auth is too blunt for agents
Traditional identity-based auth asks, “who is this?” That works for humans and stable apps, but autonomous agents need narrower authority. They need credentials that say which tool is allowed, what task it is for, how much it can spend, how long it lasts, and whether it can delegate.
SatGate puts that capability check before the request goes through. The result is not just authentication. It is rules and receipts for AI agents: Observe the call, and Control the budget and what the agent is allowed to do before the agent acts. SatGate signs a receipt when it refuses an agent over a budget, a missing payment or a tool permission, and when an agent spends from a budget in Control.
Bad default vs better default
Broad API key
One shared secret, broad account access, weak delegation, painful rotation, and no native budget boundary.
Scoped capability
A constrained token for one agent, task, route, tool set, budget, expiry window, and delegation depth.
What a capability should carry
For agent systems, authorization has to include economic policy. A safe capability answers more than who. It answers what, where, how much, how long, and who can inherit less authority.
Scope
Limit which APIs, MCP tools, routes, methods, customers, and environments an agent can access.
Budget
Attach spend caps and per-tool prices so authorization includes economic policy, not just identity.
Delegate
Allow agents to pass down narrower authority to sub-agents without handing them a master key.
Expire and revoke
Keep authority short-lived and kill risky access before the next request reaches a model, API, or tool.
Capabilities vs API keys, OAuth, and macaroons
API keys
Useful for simple server-to-server access, but usually too broad for autonomous agents and weak at scoped delegation.
OAuth
Strong for user consent and identity delegation, but often too heavyweight and human-centered for short-lived agent tool authority.
Macaroons
The token format SatGate uses for agent permissions.
Capability auth questions
What is capability-based authorization?
Capability-based authorization gives an agent a specific, constrained capability: what it can do, where it can do it, for how long, with what budget, and whether it can delegate narrower authority.
Why are capabilities useful for AI agents?
Agents act autonomously, call tools repeatedly, and delegate work. Capabilities limit blast radius by encoding scope, budget, expiry, revocation, and delegation into the authority the agent actually uses.
How are capabilities different from API keys?
API keys usually prove broad account ownership. Capabilities are narrower: they describe what this agent can access and spend right now, and they can be narrowed or revoked without rotating a shared secret.
Can capabilities include budget limits?
Yes. SatGate treats economic policy as part of authorization. A capability can carry or reference budget, per-tool pricing, route scope, tenant context, and delegation depth.
Are macaroons capability tokens?
SatGate’s tokens are macaroons, which can only be narrowed: each handoff can cut the scope, time, budget, routes and further handoffs.
Related authorization resources
Stop handing agents broad API keys.
Use capabilities that expire, narrow, meter, delegate safely, stop spend before the next request, and leave a signed receipt.
Work with SatGate