Compute an agent rents itself.

DCP is not a token pipe. It is sovereign, in-Kingdom AI compute: data, models, storage, and the control plane all stay inside Saudi Arabia under Saudi law (PDPL). One renter wallet — prepaid in Saudi Riyal — pays for everything.

An agent runs the whole lifecycle with no human and no email click. Every value below is real and verified live against api.dcp.sa.

# 1 · Self-register — no human, no email click
curl -s -X POST https://api.dcp.sa/api/renters/agent-register \
  -H "Content-Type: application/json" -d '{"label":"research-bot"}'
# → 201 { "success": true, "api_key": "dcp-renter-…", "trial_credit_sar": 20, "balance_sar": 20 }
export DCP_KEY=dcp-renter-…

# 2 · List rentable GPU TYPES (public NVIDIA labels only)
curl -s https://api.dcp.sa/api/renters/available-providers \
  -H "Authorization: Bearer $DCP_KEY"

# 3 · Rent a whole GPU — idempotent, prepaid per minute
curl -s -X POST https://api.dcp.sa/api/pods \
  -H "Authorization: Bearer $DCP_KEY" \
  -H "Idempotency-Key: $(uuidgen)" -H "Content-Type: application/json" \
  -d '{"gpu_type":"RTX 4090","duration_minutes":30}'
# → { "pod_id": "pod-…", "status": "starting" }   (or HTTP 402 with topup_url, no pod)

# 4 · OpenAI-compatible inference (qwen2.5:7b returns a clean answer)
curl -s -X POST https://api.dcp.sa/v1/chat/completions \
  -H "Authorization: Bearer $DCP_KEY" -H "Content-Type: application/json" \
  -d '{"model":"qwen2.5:7b","messages":[{"role":"user","content":"Reply with exactly: OK"}],"max_tokens":20}'
# → 200 { choices:[{ message:{ content:"OK" }, finish_reason:"stop" }], usage:{ pricing:{ sar_total:"0.0200" } } }

# 5 · Stop early — unused minutes refunded
curl -s -X DELETE https://api.dcp.sa/api/pods/$POD_ID \
  -H "Authorization: Bearer $DCP_KEY"

The official Model Context Protocol server runs over stdio via npx — nothing to install globally. It exposes eleven native tools, starting with register_agent so an agent can bootstrap its own key with DCP_API_KEY unset.

{
  "mcpServers": {
    "dcp": {
      "command": "npx",
      "args": ["-y", "github:dhnpmp-tech/dcp-mcp"],
      "env": { "DCP_API_KEY": "dcp-renter-…" }
    }
  }
}

Base URL https://api.dcp.sa/v1. GET /v1/models returns OpenAI-style entries, each with an available flag — call only models with available=true. Each completion carries per-call usage pricing in both USD and SAR.

from openai import OpenAI
client = OpenAI(base_url="https://api.dcp.sa/v1", api_key="dcp-renter-…")

resp = client.chat.completions.create(
  model="qwen2.5:7b",  # GET /v1/models → pick one with available=true
  messages=[{"role": "user", "content": "اشرح لي زكاة المال"}],
)
print(resp.choices[0].message.content)

On-demand types spin up a whole, dedicated NVIDIA GPU in about a minute. Rates are cost-plus from the live market — each is a 'from' floor that floats. The native RTX 3090 is an in-Kingdom community card. USD ≈ SAR ÷ 3.75; billing is in SAR.

Live availability per type: GET https://api.dcp.sa/api/renters/available-providers. Full per-token and subscription pricing lives on the pricing page. See pricing →

Two machine-readable guarantees make money safe for autonomous callers: an Idempotency-Key so a retry never double-spends, and an HTTP 402 funding signal that says exactly how much is needed and where to top up — and creates no pod or charge.

# POST /api/pods H100 / 600 min over a 20 SAR wallet → no pod created
{
  "error": "insufficient_balance",
  "code": "insufficient_balance",
  "currency": "SAR",
  "required_sar": 172.73,
  "balance_sar": 20,
  "topup_url": "https://dcp.sa/renter/wallet",
  "retryable": true
}

Inference, GPU pods, and persistent volumes all run on Saudi-owned hardware inside the Kingdom under full PDPL data-residency. Cross-border frontier models are off unless a tenant explicitly opts in — and when they do, every such request is marked. This is the structural advantage no foreign API can match for KSA and Gulf workloads.