Temporal MCP Server

⚠️ Unofficial Project · This is an independent, community-built MCP server and is not affiliated with, endorsed by, or officially supported by Temporal Technologies, Inc.

An MCP (Model Context Protocol) server for the Temporal HTTP API. Instead of jumping between the Temporal Web UI, tctl/temporal CLI, and your editor every time you need to inspect a workflow, you just ask your AI: “List failed workflows in the last hour,” “Describe workflow order-42 and show me the history,” or “Signal cart-17 with checkout.” The server talks to your Temporal cluster over the HTTP API and runs the right operations — from Cursor, Claude Desktop, Claude Code, or any MCP-compatible client.

What is it?

A bridge between your AI assistant and a Temporal cluster. It exposes 36 tools covering the full Temporal HTTP API surface — cluster info, namespaces, workflows, schedules, activities, batch operations, worker deployments, Nexus endpoints, workflow rules, task queues, and search attributes.

Because loading all 36 tools into the LLM at once bloats the context, the server ships with a tiered tool system: by default only the 11 most-used tools are exposed, and you opt into more when you need them.

Key Features

• 36 tools, 3 tiers: essential (11, default), standard (23), or all (36) — set with one env var.
• Full workflow lifecycle: list, describe, start, signal, signal-with-start, query, cancel, terminate, pause/unpause, and full history with human-readable summaries.
• Schedules: create, list, describe, and delete cron/interval schedules.
• Activities & task queues: inspect activity executions, heartbeats, failures, and active pollers.
• Advanced ops (in all): batch operations, versioned worker deployments, Nexus endpoints, and workflow rules (auto TERMINATE/PAUSE by visibility query).
• Structured output: key tools return both a human-readable summary and a machine-readable structuredContent JSON — clients that support MCP structured output (like Claude) can process results without text parsing.
• Authenticated clusters: supports Temporal Cloud and self-hosted clusters via TEMPORAL_API_KEY.
• NPX-ready: no global install needed.

Quick Start

1. Add to Cursor MCP config (.cursor/mcp.json in your project, or ~/.cursor/mcp.json globally):

{
  "mcpServers": {
    "temporal": {
      "command": "npx",
      "args": ["-y", "@alisaitteke/temporal-mcp"],
      "env": {
        "TEMPORAL_ADDRESS": "http://localhost:8080",
        "TEMPORAL_NAMESPACE": "default",
        "TEMPORAL_TOOLS": "essential"
      }
    }
  }
}

2. Restart Cursor. The Temporal tools will appear in the MCP tool list.

3. Ask your AI:

"List the 10 most recent failed workflows"
"Describe workflow order-42 and show the last 20 history events"
"Signal workflow cart-17 with checkout, payload { coupon: 'SUMMER' }"
"Create a daily schedule at 03:00 UTC that starts the ReportWorkflow"

Environment Variables

Tool Tiers

Pick the tier that matches your job: essential for day-to-day debugging, standard for scheduled/cron-heavy workloads, all for platform/operator work.

Technologies

• TypeScript
• MCP (Model Context Protocol) SDK
• Temporal HTTP API
• Zod (schema validation)

Installation

# Run with npx (recommended)
npx -y @alisaitteke/temporal-mcp

# Or install globally
npm install -g @alisaitteke/temporal-mcp

Requires Node.js ≥18 and a reachable Temporal server (local dev server, self-hosted, or Temporal Cloud).

Links

• 📦 NPM Package
• 💻 GitHub Repository
• 📖 Temporal HTTP API

If you’ve ever had to explain a stuck workflow over Slack at 2 AM, this is the tool that lets your AI pull the history, describe the activity, and hand you the failure reason — without leaving the editor.