SuperClaw / OpenCode (MCP)

SuperClaw is Intel's hybrid, local-first agentic AI, built on OpenCode. Like other OpenCode-based agents, it is MCP-extensible — so its agents can call ColabHive's platform (run inference, train on your data, deploy models) as native tools, alongside the local-first work it does on the AI PC.

This complements pointing SuperClaw's cloud model at ColabHive (see Use ColabHive with SuperClaw / OpenCode): the cloud model does the reasoning, and these MCP tools let the agent act on the ColabHive platform.

Add the ColabHive MCP server

Use the hosted endpoint with your API key (see Authentication):

{
  "mcp": {
    "colabhive": {
      "type": "remote",
      "url": "https://mcp.colabhive.com/mcp",
      "headers": { "Authorization": "Bearer hive_..." }
    }
  }
}

For OpenCode / Claude Code the shape is the same as any MCP server — see Custom client setup for the local (stdio) variant.

What the agent can do

Once connected, the agent gets ColabHive's tools, including:

Tool	Purpose
list_endpoints	Discover invocable models/endpoints (base models + your deployed ones) with readiness
get_endpoint	Endpoint details + warm/cached/cold readiness
run_inference	Run sync inference on an endpoint — LLMs (incl. tool-calling), specialists, etc.
list_trainable_models / get_model_schema	Discover base models to fine-tune and their hyperparameters
create_dataset / list_datasets	Register and list datasets
create_training_run / get_training_run / get_training_metrics / get_training_logs	Fine-tune on your proprietary data and track it
register_for_inference	Deploy a finished training run as an invocable endpoint
list_models	List your trained/registered models

Trained endpoints also surface automatically as per-model inference tools.

Why this fits SuperClaw

• Sovereign + Intel-native. ColabHive serves on Intel Arc, so the heavy work stays on Intel silicon, edge-to-cloud, matching SuperClaw's privacy-first design.
• Act on proprietary data. The agent can fine-tune and run specialists (embeddings, rerank, forecasting) on the customer's own data — the RAG + agentic-workflow story, executed on the platform.

tip

For tool-calling with a cloud model (the model emitting tool_calls that SuperClaw executes locally), see Tool Calling. MCP is the complementary path where the agent drives the ColabHive platform itself.