The task manager for Codex CLI
Give OpenAI's Codex CLI — the terminal coding agent — a real task queue. TaskPrio is MCP-native: Codex pulls the top task with get_next_task, edits files and runs commands in your terminal, reports a result, and grabs the next. One priority order, worked top to bottom — watched live in a cockpit.
Add it to Codex
First fetch the bundled server and verify your token (one-time — grab the token from the TaskPrio sidebar → Copy token):
PROMPTPRIO_API_TOKEN=pp_xxx bash <(curl -fsSL https://taskprio.com/install.sh)
That caches the MCP server to ~/.local/share/promptprio/mcp-server.js. Now register it with Codex. Codex configures MCP servers in ~/.codex/config.toml — add an [mcp_servers.promptprio] table:
[mcp_servers.promptprio]
command = "node"
args = ["/Users/you/.local/share/promptprio/mcp-server.js"]
env = { PROMPTPRIO_API_TOKEN = "pp_your_token", PROMPTPRIO_API_URL = "https://taskprio.com" }
Prefer the CLI? codex mcp add writes the same entry. Either way, on the next run Codex loads the TaskPrio tools. Full steps in the docs.
The autopilot loop
1. get_next_task → top unblocked task + merged project context
2. (Codex edits files + runs commands in your terminal)
3. complete_task → short result, logged for the audit trail
4. repeat → until the queue returns no task
To run unattended, start Codex in full-auto (approval set to never) so it clears the queue without a confirmation between tasks. Each handed-out task is leased, so Codex can run beside a Claude Code, Cursor or Cline session on the same queue without collisions.
Why developers pair Codex with TaskPrio
- No prompt babysitting — Codex pulls the next task itself instead of you typing a new instruction after each completion.
- One source of truth — the same ranked queue you watch on the web cockpit is the one Codex reads.
- Context travels with the task — per-project instructions, working directory and definition-of-done are merged into every task Codex receives.
- You set priority — capture into the Inbox, rank what matters; Codex only works ranked tasks, never an unfiltered pile.
- Honest by design — every completion records a result, so an unattended full-auto run leaves an audit trail you can scan later.
Frequently asked questions
How do I add a task queue to the Codex CLI?
Run the one-line installer, then add an [mcp_servers.promptprio] block to ~/.codex/config.toml (or run codex mcp add). Codex then has get_next_task and complete_task.
Can it run unattended?
Yes — start Codex in full-auto so it loops get_next_task → work → complete_task on your queue without a confirmation per step, the same autopilot loop Claude Code and Cursor use.
Can I run it next to Claude Code?
Yes. Tasks are leased on hand-out, so multiple agents pull from one priority order without grabbing the same task.
Is it free?
Yes. Free includes 1 MCP session and 50 completions/day; Pro ($10/mo) gives 5 sessions and unlimited completions — no per-agent fees, no extra tokens.
Start free — no card → Using Claude Code? → Cline? →
Run it on autopilot. Paste our maintained autopilot prompt into Codex — it adds hard safety gates (money / credentials / publishing stay manual), an empty-queue stop, and auto-scopes to your project.
view ↗Done for now? Paste the stop prompt to end the session cleanly — it lands the current task (done-with-proof or a resume note), syncs the queue true, banks one learning, and leaves you a short report.
view ↗