Cron and webhooks
A task fires on a schedule, or when a POST arrives at its webhook, and the JSON body of that request can be placed in the prompt.
task_server
Scheduled AI tasks: a prompt, a trigger, and every run kept.
What it does
A task is a prompt. Give it a trigger and the server runs it by itself, against any OpenAI-compatible model, with the tools of every MCP server you have connected. There is no conversation here, only tasks that fire on their own, and a record of what each one did. One process serves the web app, a GraphQL API, and the same API as MCP tools for other agents.
-
-
Your model, your tools
Point it at Ollama, LM Studio, OpenAI or OpenRouter, add stdio or HTTP MCP servers, and every run can reach their tools.
-
Every run kept
Status, output, tool calls and token counts are stored for each run, and a run in flight can be watched as it happens.
Get started
git clone https://github.com/cubicecho/task_server.git
cd task_server
docker compose up --buildOpen http://localhost:8787, go to Settings, point the base URL at an OpenAI-compatible server and pick a model. Then create a task, give it a cron expression, and press Run now. The database is embedded, so there is nothing else to install.
Full documentation lives in the project README: https://github.com/cubicecho/task_server