@sensigo/realm-mcp

@sensigo/realm-mcp — the Realm MCP server. Exposes 7 workflow tools over stdio or HTTP for AI agent connections (VS Code Copilot, Cursor, Claude, and any MCP-compatible agent).

Installation

# Standalone binary (for AI agent MCP config)
npm install -g @sensigo/realm-mcp

# Embedded library (for custom application integration)
npm install @sensigo/realm-mcp

Usage — Standalone MCP server

Add realm-mcp to your MCP client configuration (VS Code, Claude Desktop, Cursor, etc.):

{
  "mcpServers": {
    "realm": {
      "command": "realm-mcp"
    }
  }
}

Requires workflows to be registered first via realm workflow register.

Usage — Embedded MCP server

Create and connect a Realm MCP server inside your own application using any MCP-compatible transport.

import { createRealmMcpServer } from '@sensigo/realm-mcp';
// StdioServerTransport comes from the MCP SDK, not from @sensigo/realm-mcp
import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';

const server = createRealmMcpServer(); // options?: RealmMcpServerOptions
const transport = new StdioServerTransport();
await server.connect(transport);

API Reference

| Symbol | Description | | ---------------------------------- | ------------------------------------------------------------------------------- | | createRealmMcpServer(options?) | Creates the MCP server with all 7 tools pre-registered. Returns McpServer. | | createDefaultRegistry() | Returns an ExtensionRegistry pre-populated with built-in adapters. | | generateProtocol(workflow) | Generates a structured protocol description for a workflow. | | RealmMcpServerOptions | Type — optional config: registry?, secrets?, workflowStore?, runStore?. | | WorkflowProtocol, ProtocolStep | Types — output shape of generateProtocol. |

MCP tools exposed

• list_workflows — list registered workflows
• get_workflow_protocol — get step-by-step protocol for a workflow
• start_run — start a new workflow run
• execute_step — submit agent output for a step and advance the run
• submit_human_response — resolve a human gate
• get_run_state — check current run state
• create_workflow — dynamically register and start a workflow in one call

Full documentation

Full documentation: https://github.com/sensigo-hq/realm