Skip to main content

What are contractors?

Contractors let your Claude agents delegate tasks to other AI models. A contractor is a connection to an external model provider (OpenRouter, OpenAI, etc.) that agents can call via MCP tools during a run. Think of it as hiring a specialist. Your Claude agent stays in charge but can ask a GPT-4o or Grok instance for a second opinion, a different perspective, or a task that another model handles better.

How it works

1

Sign an agreement

On the Contractors page, create a contractor by providing a name, API provider, API key, model ID, and base URL. This is the “contract.”
2

Agent calls the contractor

During a run, the agent uses the ask_contractor MCP tool with the contractor’s name and a prompt. The MCP server validates the contract is active, then calls the external API.
3

Response flows back

The contractor’s response is returned to the agent as a tool result. The agent can use it alongside its own findings.

MCP tools

The mcp-contractor server provides two tools:
ToolDescription
ask_contractorSend a prompt to a named contractor and get back their response
list_contractorsDiscover which contractors are currently active

Managing contractors

The Contractors page shows all contracts as cards. Each card displays:
  • Provider and model
  • Status (active or inactive)
  • Toggle to activate/deactivate
  • Terminate button to permanently end the contract
Contractor API keys are stored securely. They are never returned on public API requests. Only internal MCP server calls (authenticated via the internal secret) can access them.

Compatible providers

Any provider with an OpenRouter-compatible chat completions API works:
  • OpenRouter (access to hundreds of models)
  • OpenAI (GPT-4o, GPT-4, etc.)
  • xAI (Grok)
  • Google (Gemini, via OpenRouter)
  • Any custom endpoint that implements the /chat/completions format