← All docs

Compatibility & badges

Sanction fits where agents already run. It is not another agent framework; it is the authorization plane those frameworks can call before an agent spends money, invokes a tool, touches a credential, provisions a resource, or burns through a model budget.

Badges

Use these claims when an integration actually routes through the named Sanction surface:

BadgeUse whenProof surface
Sanction-governed MCPAn MCP host or server calls sanction-mcp before risky tools or spendnpx sanction-mcp, /authorize/tool, /authorize
AuthZEN PDP compatibleA policy enforcement point calls Sanction's OpenID AuthZEN endpoints/access/v1/evaluation, /access/v1/evaluations
AARP approval loopAn AuthZEN escalation opens a Sanction access request and redeems the grant on retry/access/v1/access-request
Gateway meteredModel calls route through the Sanction gateway with x-sanction-key/api/gateway/<provider>
Evidence replay readyDecisions persist policy revision + context and can replay the pure rules/authorize/{id}/evidence

MCP hosts

The fastest channel is MCP because the install shape is already familiar:

{
  "mcpServers": {
    "sanction": {
      "command": "npx",
      "args": ["sanction-mcp"],
      "env": {
        "SANCTION_API_KEY": "pxy_...",
        "SANCTION_WALLET_ID": "wal_..."
      }
    }
  }
}

Use the MCP tool governance policy pack when the first job is tool control: read/search tools can pass, writes and shell commands escalate, delete/payment tools deny.

Frameworks

Framework integrations should keep their native developer experience and add one pre-action check:

  1. Route model calls through /api/gateway/<provider> when token budgets matter.
  2. Call /api/v1/authorize/tool before tool invocations that can change state.
  3. Call /api/v1/authorize before purchases or subscriptions.
  4. On escalated, wait for the grant and retry with grant_id.
  5. Link the resulting decision/evidence id into the framework's trace.

Start with the Coding agent seat pack for coding tools, or Client-safe launch for agency-delivered agents.

Gateways and proxies

If a team already uses LiteLLM, Portkey, Helicone, Envoy AI Gateway, Cloudflare AI Gateway, or an internal proxy, Sanction does not need to replace it on day one. Treat Sanction as the external policy authority:

  • Meter directly through the Sanction gateway for the simplest pilot.
  • Or keep the existing gateway and call Sanction's REST/AuthZEN endpoints before the gateway forwards risky actions.

Use the Gateway token budget pack for the first pilot: model calls are metered with a hard daily token budget, while broader spend/tool governance can come next.

Payments and mandates

Payment standards such as AP2 and x402 are rails and mandate formats. Sanction's role is policy and evidence before a rail is used: who authorized the agent, what limits applied, which human approved the exception, and which grant was redeemed.

Use the Payment agent mandate pack when every money movement should escalate and the evidence trail matters more than speed.

Sanction Local (air-gapped)

Regulated practices that keep models on their own hardware need a policy that denies cloud egress and an export an assessor can verify without a walkthrough.

Use the No egress pack (no-egress): only exact on-box tools (local.ollama, local.chroma, local.embeddings, local.memory) are allowed; named cloud tools are blocked; anything else on the allow-list path is denied and persisted. Capability installs escalate or block. Pair it with the signed audit export (GET /v1/audit/export or Signed evidence on the dashboard Audit page) and POST /v1/audit/verify.

Sanction decides and evidences; the runtime must still refuse egress. The pack is the governance half of the Local install package.

Start free →