AI Features

gocron ships a set of AI capabilities, all powered by your own OpenAI-compatible LLM (cloud or local): natural-language-to-cron, failure-log diagnosis, an AI ops chat assistant, and MCP protocol integration.

Prerequisite: configure the LLM

Go to System → AI Config, fill in and enable:

Field	Description
Enable	AI features are available only when enabled
Base URL	OpenAI-compatible endpoint, e.g. https://api.openai.com/v1 or local http://localhost:8000/v1
API Key	Access key — stored server-side only, never returned to the frontend
Model	Model name, e.g. gpt-4o, qwen2.5

Any OpenAI-compatible endpoint works: a cloud API, or a local model server (Ollama, vLLM, LM Studio, MLX, …). A local model gives you fully offline AI.

Note: "thinking/reasoning" models are slower on multi-step tasks — this is expected; the chat panel streams the reasoning live and you can stop it anytime.

Natural-language to Cron

When creating / editing a task, describe the schedule in natural language (e.g. "every day at 9:30 am", "hourly on weekdays"). The AI generates gocron's 6-field second-level cron expression and previews the next run times for you to confirm.

AI failure-log diagnosis

After a task run fails, click Diagnose on the task log. Based on the task config and the failed output, the AI returns:

• a one-sentence root cause
• actionable fix suggestions

AI ops assistant (chat)

Click the robot icon in the top bar to open the chat panel and ask ops questions in natural language:

• "Which tasks failed recently?" / "Why did task X fail?" / "What task templates do I have?"
• Streaming output, with the model's reasoning shown live
• The AI calls read-only tools to fetch real data (tasks, logs, hosts, templates) and answers from the results — it does not fabricate
• Triggering a task run (run_task) requires your explicit confirmation and is audited — the AI never runs a task automatically
• Supports Markdown rendering, message copy, a resizable panel, and stop-anytime
• Reasoning and answers follow your UI language

MCP integration

gocron has a built-in MCP (Model Context Protocol) server, so external MCP clients (e.g. Claude Desktop) can connect and use gocron as a tool to manage tasks.

• Endpoint: https://<your-host>/mcp
• Create an access token under System → MCP Keys (the token is shown in full only once at creation — save it)
• Client config example:

{
  "mcpServers": {
    "gocron": {
      "url": "https://<your-host>/mcp",
      "headers": {
        "Authorization": "Bearer <your MCP token>"
      }
    }
  }
}

Available tools:

Tool	Description
list_tasks	List tasks (filter by name / tag / status)
get_task	Get a single task's details
query_task_logs	Query execution logs (filter by task / status / keyword / time range)
list_hosts	List execution nodes
list_templates	List task templates
diagnose_task_log	Diagnose a failed log
run_task	Run a task now (requires an admin token)

Security: run_task is the only write operation and requires an admin token; the rest are read-only. Expose the /mcp endpoint over TLS (HTTPS).