ArtiBot Cloud MCP - Zhijiangyun 智匠云認証済み注目

What is ArtiBot Cloud MCP?

ArtiBot Cloud MCP (Zhijiangyun 智匠云) is a cloud MCP server for robot manipulation data and GPU workloads. It exposes 16 tools over Streamable HTTP: robot fleet catalog, task templates, data collection & video recording, model training, inference management, and benchmark evaluation — all under a unified permission model where tool access is identical to the web UI (same plan quotas, GPU time accounting, and admin roles).

Endpoint: https://mcp.artibot.cloud/mcp · Transport: Streamable HTTP (MCP 2025-03-26+) · Auth: Authorization: Bearer zj_ext_... (account-bound API key) · Rate limit: 30 req/min per client, plus 30 req/min per IP at the gateway.

tools/list is filtered by your account automatically: a normal account sees the public catalog, read-only platform tools, and GPU workload tools; infrastructure tools (Docker/environment) are only visible to real admin accounts; keys not bound to an account can only use the public catalog.

Tools

• get_service_status - Public service status of the ArtiBot Cloud MCP (release stage, transport, auth mode, permission model). Reads no project data, creates nothing, bills nothing.
• list_public_capabilities - Lists currently enabled public MCP tools and the minimal scope required for each. Never exposes internal admin tools.
• list_public_task_templates - Lists reviewed, publicly visible task templates (task name, Chinese title, category, applicable robots). Use a template name from here as taskName for start_data_collection.
• list_robots - Lists the robots visible to your account: official models plus custom robots adapted through the generic pipeline. Read-only.
• get_checkpoints - Lists model checkpoints produced by your training runs (your own data only). Read-only.
• get_inference_status - Shows the runtime status of your model inference service. Read-only.
• get_eval_sessions - Lists your historical evaluation sessions. Read-only.
• get_eval_scores - Shows your model evaluation scores, with dual-path operation/cognitive metrics. Read-only.
• list_recorded_videos - Lists your recorded video catalog (your own data only). Read-only.
• get_recording_video_url - Returns an access URL for one of your recorded videos. Read-only.
• network_precheck - Runs a network connectivity precheck without starting any workload.
• start_data_collection - Starts an automated data collection run from a task template (dedicated simulation environment + recording). Consumes your plan's GPU time with the same quota validation as the web UI; output is registered as your private data asset.
• start_training - Starts imitation-learning / reinforcement-learning training under your account. Consumes subscribed GPU time and budget with the exact same billing & quota validation as launching from the web UI.
• manage_inference - Starts, stops, and manages inference services under your account. Consumes subscribed resources with the same quota checks as the web UI.
• run_benchmark - Runs a benchmark evaluation task under your account. Consumes subscribed resources with the same quota checks as the web UI.
• cancel_own_workload - Cancels one of your workloads (by commandTaskId). Shares the web UI's ownership validation, billing stop, and ACK resource-release pipeline.

Grouping: the first three tools form the public catalog (capabilities:read, no account data, no billing); the next eight are read-only platform queries over your own account data (platform:read); the last five are GPU workloads (platform:write, billed against your plan with the same quota checks as the web UI).

How to use

1. Get your API key at https://mcp.artibot.cloud/keys (binds to your ArtiBot Cloud account; self-issued, max 5 per IP per day).
2. Add the server config (see Config tab) to any MCP client: Claude Desktop, Cursor, Cline, WorkBuddy, etc.
3. Ask in natural language, or call tools directly — e.g. list_robots, list_public_task_templates, start_data_collection, list_recorded_videos.

{
  "mcpServers": {
    "artibot-cloud": {
      "type": "streamableHttp",
      "url": "https://mcp.artibot.cloud/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_ARTIBOT_API_KEY>"
      }
    }
  }
}

Key features

• 16 tools across public catalog, read-only platform queries, and GPU workloads (data collection, training, inference, benchmark).
• Unified permission model: web UI, in-platform agents, and external MCP clients share exactly the same permission rules.
• Plan quotas & GPU time accounting built in; billable tools run the same quota validation as the web UI.
• Full audit trail: every call (including rejected ones) is logged and can be requested back.
• Works with any MCP-compatible client (Streamable HTTP + Bearer auth).

Limits & error codes

• Rate limit: 30 req/min per client (429 RATE_LIMITED when exceeded); gateway additionally limits 30 req/min per IP.
• Scopes: capabilities:read (public catalog) · platform:read (read-only platform tools) · platform:write (GPU workload tools).
• Error codes: 401 MISSING_OR_INVALID_API_CLIENT / 401 INVALID_OR_EXPIRED_API_CLIENT / 403 INSUFFICIENT_SCOPE / 403 IP_NOT_ALLOWED / 429 RATE_LIMITED / 503 PUBLIC_MCP_DISABLED. Every error response carries a request_id — include it when reporting issues.

Links

• Platform: https://www.artibot.cloud
• MCP docs: https://mcp.artibot.cloud
• Health check (no auth): https://mcp.artibot.cloud/health