Skip to main content

Model Context Protocol (MCP)

Arrotech Hub is built on the Model Context Protocol (MCP), an open-standard architecture that allows AI models to securely interact with local and remote data sources, tools, and APIs.

Why MCP?​

Traditional AI integrations are static and hard-coded. MCP allows Arrotech to:

  • Hot-swap Tools: Add new integrations (e.g., a new CRM) without rebooting the entire AI service.
  • Dynamic Discovery: Your AI agent can "query" the hub to see what tools are available in real-time.
  • Provider Agnostic: Use any LLM (OpenAI, Anthropic, Ollama) with the same set of tools.

The Tool Registry​

The Tool Registry is the heart of our MCP implementation. It manages:

1. Tool Definitions​

Each tool is defined by a name, a description, and a JSON Schema that dictates the required input parameters.

2. Runtime Execution​

When you call /mcp/call, the registry:

  1. Locates the tool implementation.
  2. Validates your input against the JSON Schema.
  3. Injects your encrypted Connections (tokens).
  4. Executes the API request and returns a structured response.

3. Tool Discovery​

The /mcp/tools endpoint provides a catalog of every tool your workspace is authorized to use.

API Integration​

For custom tool development or manual execution, use these endpoints:

tip

Always use the include_all=true flag on the /tools endpoint during development to discover tools that may be inactive due to missing connections.