AI agents & automation
Two ways to let AI agents act on the platform: scoped Agent Credentials (agcred_) for delegated wallet transfers and payment links, and the MCP server for LLM tool-call integration with campaigns, slots, and wallets.
The platform offers two distinct systems for integrating AI and automation. Choose the one that fits your use case—or use both.
System 1 - Agent Credentials (scoped delegation)
What it does: You issue a scoped credential key (agcred_...) to your AI agent or automation script. The agent calls a small set of money-movement endpoints on your behalf, within limits you set.
When to use: Your AI or script needs to *move money*—send a wallet transfer or create a payment link—on behalf of your account, with hard guardrails. You set a per-transaction limit and a daily cap; the platform enforces them atomically and the agent cannot exceed them.
| Feature | Detail |
|---|---|
| Auth | Authorization: Bearer agcred_<key> |
| Endpoints | POST /agent/transfer, GET /agent/balance, POST /agent/payment-link |
| Per-txn limit | maxAmountPerTxn, set when issuing the key |
| Daily cap | maxAmountPerDay, atomically enforced; auto-decremented on error |
| Recipient scope | Optional recipientWhitelist (E.164 phone numbers) |
| Callbacks | Optional HTTPS webhookUrl / per-call callbackUrl override |
| Idempotency | Pass idempotencyKey to prevent duplicate transfers |
| Audit log | Every call recorded per credential (GET /agent-me/:id/transactions) |
→
System 2 - MCP server (LLM tool calls)
What it does: Connect any MCP-compatible AI model (Claude, GPT, Gemini, Llama, Mistral…) to the platform's campaign, slot, wallet, and trust tools. The agent discovers available tools via a manifest endpoint, then calls them using standard JSON-RPC over Streamable HTTP.
When to use: Your LLM needs to *orchestrate* a broader workflow—discover campaigns, reserve and submit slots, check earnings, inspect wallet balances, manage Relay rules, check trust scores—through self-describing tool calls without custom integration code.
| Feature | Detail |
|---|---|
| Auth | User JWT or automation OAuth access token (remote connectors) + mcp_server entitlement |
| Discovery | GET /.well-known/mcp.json |
| Transport | Streamable HTTP (GET/POST/DELETE /mcp) |
| Tools | ActionRegistry gateway tools (campaign, slot, platform, crypto checkout, payout) plus supplemental Trust, Relay, Insights, and agent_check_balance. Delegated transfers and Clip Studio are not MCP tools. |
| Errors | Structured ApiErrorCode for deterministic agent decisions |
| Audit | Every tool call logged with agent ID, timestamp, inputs, outputs |
→
Choosing between the two
Agent Credentials (agcred_) | MCP server | |
|---|---|---|
| Auth key type | agcred_ scoped key | User JWT or automation OAuth token; both need mcp_server entitlement |
| What the AI can do | Wallet transfers + payment-link creation, within set limits | Advertiser create/list/fund (confirm before money) and creator discover/reserve/submit/earnings, plus wallet, Relay, and trust reads |
| Spend guardrails | Per-txn limit + daily cap on the credential | OAuth tokens are scope-limited. Money tools are marked destructive; confirm with the user before calling them. |
| Idempotency | Built-in via idempotencyKey field | Idempotency-Key header |
| Callbacks | HTTPS webhookUrl/callbackUrl | Streamable HTTP session |
| Best for | Delegated, tightly-scoped money movement | Broader agentic campaign/creator workflows |
Related
- - event-driven integrations (Zapier/Make/n8n)
- Actions Gateway OpenAPI (GPT Actions):
GET https://api.ugc.africa/actions-gateway/openapi.json· interactive docs/actions-gateway/docs
Did this solve your problem?