dcp-mcp

@dcp/mcp — DCP Model Context Protocol server

DCP is Saudi Arabia's sovereign AI compute platform — in-Kingdom inference, GPU rental, and persistent storage, billed in SAR, on Saudi-owned hardware. This MCP server lets any MCP-capable agent (Claude Desktop, Claude Code, Cursor, or your own) use DCP through native tool calls.

It is a thin client: it reads DCP_API_KEY from the environment and calls https://api.dcp.sa. No secrets are bundled.

Install (one line)

The connector runs straight from GitHub — no npm install, no clone:

DCP_API_KEY=dc1-sk-xxxxxxxx npx -y github:dhnpmp-tech/dcp-mcp

MCP client config

Add this to your MCP client (Claude Desktop claude_desktop_config.json, Claude Code / Cursor .mcp.json):

{
  "mcpServers": {
    "dcp": {
      "command": "npx",
      "args": ["-y", "github:dhnpmp-tech/dcp-mcp"],
      "env": { "DCP_API_KEY": "dc1-sk-xxxxxxxx" }
    }
  }
}

Then ask your agent to "list DCP models" or "rent a GPU on DCP for 30 minutes." Agents with no key can call register_agent first — see the quickstart below.

Tools

Tool	What it does
register_agent	Self-register in one call — get a real API key + small SAR trial, no human, no email click. Use first when you have no key.
list_models	List currently-serveable models (OpenAI-style).
chat	Run a chat completion (OpenAI-compatible, in-Kingdom).
get_balance	Renter wallet balance (SAR).
list_gpus	List rentable GPU types (e.g. H100, RTX 4090) with VRAM + live availability.
create_pod	Rent a whole GPU as an interactive pod (root + Jupyter + SSH), prepaid per minute.
get_pod	Pod status + access URL / SSH command / time remaining.
extend_pod	Add time to a running pod, no restart.
stop_pod	Stop early; unused prepaid time refunded.
rent_volume	Rent an exclusive in-Kingdom persistent volume (10/20/30 GB).
get_volume	Active volume status + usage.

Zero-human quickstart (for agents)

The whole loop runs with no human in it:

1. register_agent → returns a real api_key + a 20 SAR trial credit (no email click). Set it as DCP_API_KEY.
2. list_gpus → pick a gpu_type string (e.g. "H100", "RTX 4090") from the live, available types.
3. create_pod with that gpu_type + duration_minutes → poll get_pod for the access_url / ssh_command once running.
4. chat → run OpenAI-compatible inference on an available model from list_models.
5. stop_pod → stop early; unused prepaid minutes are refunded to the wallet.

Minting a key by hand (equivalent to register_agent):

curl -s -X POST https://api.dcp.sa/api/renters/agent-register \
  -H 'Content-Type: application/json' -d '{}'
# → { "api_key": "dcp-renter-…", "trial_credit_sar": 20, "balance_sar": 20, ... }

The trial (20 SAR) is enough to list GPUs, run a short pod, and do real inference; the larger grant stays behind email-verified signup. Calls are per-IP rate-limited.

Environment

• DCP_API_KEY — renter API key (required for every tool except register_agent).
• DCP_API_BASE — API host, default https://api.dcp.sa.

Why DCP

Inference, GPU rental, fine-tune hosting, and storage on Saudi-owned hardware inside the Kingdom — full PDPL / data-residency compliance, billed in SAR. The inference API is a drop-in OpenAI replacement: point any OpenAI SDK at https://api.dcp.sa/v1.

Learn more: https://dcp.sa/v2/agents · https://dcp.sa/llms.txt

MIT licensed.