> ## Documentation Index
> Fetch the complete documentation index at: https://docs.useagentbank.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Architecture

> Understand the responsibility boundaries between the user, agent, MCP, Core, wallet, and payment rails.

```mermaid theme={null}
flowchart LR
    U["User"] --> A["AI agent"]
    A --> M["AgentBank MCP"]
    M --> C["AgentBank Protocol Core"]
    C --> Q["Quote and payment services"]
    C --> W["Privy wallet infrastructure"]
    C --> R["Payment rails and settlement partners"]
    R --> B["Bank or local payment destination"]
    U --> P["AgentBank browser action"]
    P --> C
    P --> I["Privy, World ID, KYC, or fiat action"]
```

| Layer | Responsibility |
| - | - |
| User | Connect the account, set the approval threshold, and complete required browser actions |
| AI agent | Understand the request, call MCP tools, present material details, and track status |
| AgentBank MCP | Restore a local installation or authenticate a hosted OAuth connection, then expose its high-level tools and resources |
| Protocol Core | Enforce authorization, lock routes, issue instructions, verify execution, and store durable state |
| Privy | Provide the AgentBank wallet infrastructure |
| World ID / AgentKit | Provide payment-specific human authorization and verified-agent capabilities |
| KYC provider | Verify eligibility for gated fiat markets |
| Payment rails | Collect or deliver fiat for supported routes |

<Warning>
  A successful on-chain receipt is not necessarily final payment completion.
  Always use `get_payment` as the authoritative durable status.
</Warning>