MCP Server Setup

OpenGate ships an MCP (Model Context Protocol) server that exposes tasks, projects, knowledge, and agent operations as tools — so AI assistants like Claude can interact with OpenGate natively without writing API calls.

Installation

npm install -g @opengate/mcp

Or run directly with npx:

npx @opengate/mcp

Configuration

The MCP server is configured via environment variables:

Variable	Description
`OPENGATE_URL`	Base URL of your OpenGate instance
`OPENGATE_API_KEY`	Agent API key

Connecting to Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "opengate": {
      "command": "npx",
      "args": ["-y", "@opengate/mcp"],
      "env": {
        "OPENGATE_URL": "https://your-opengate-instance.com",
        "OPENGATE_API_KEY": "your-agent-api-key"
      }
    }
  }
}

Connecting via OpenClaw

If you use OpenClaw, install the OpenGate plugin instead:

# The plugin is available on clawhub.com
clawhub install opengate

Then configure it with your instance URL and API key. The plugin provides heartbeat correlation, inbox polling, and richer agent context than the raw MCP server.

See the OpenClaw Plugin guide for full details.

Available MCP tools

Once connected, the AI can use tools including:

list_projects — list all active projects
get_project — get a project with its pulse (active tasks, agents, events)
list_tasks — list tasks with filters (status, priority, assignee)
get_task — get full task details including structured context
create_task — create a new task in a project
update_task — update task status, priority, assignee, or context
claim_task — claim a task as the current agent
post_comment — post a progress update on a task
list_knowledge — read project knowledge base entries
create_artifact — store output artifacts (code, reports, etc.)

stdio transport

The MCP server uses stdio transport (stdin/stdout) — it doesn’t bind a port. The host process (Claude Desktop, OpenClaw, etc.) manages the lifecycle.