Give your coding agent a to-do list that survives the session

A coding agent can use World's Best To-Do List (WBTDL) through MCP to save next actions between sessions. Connect a compatible client, have it read the existing tasks, and update the relevant task when the work is checked. Your task list remains in WBTDL even when the coding conversation ends.

This gives “I'll fix that next” a place to live. It doesn't save the entire debugging session or replace the repository's documentation.

Example: connect Codex CLI

OpenAI's Codex MCP documentation covers remote servers and OAuth. With Codex CLI installed, add the server and start sign-in:

codex mcp add wbtdl --url https://worldsbesttodolist.com/api/mcp
codex mcp login wbtdl

Complete the WBTDL sign-in flow. In a Codex CLI session, use /mcp to inspect the connection, then ask the agent to read your pages and categories without making changes. Adding the server configuration alone doesn't verify account access.

For another client, use its remote MCP and OAuth instructions with the same endpoint. The WBTDL connection page links to the conversational assistant guides.

Start with the saved task and the current code

Read my open WBTDL tasks in the Website category. Find “Show a clear error when a card payment fails.” Inspect this repository and propose the change before editing. Keep the task open until the agreed checks pass.

The title identifies the work. The repository supplies the implementation context. Give the agent the relevant issue, acceptance criteria, or document if the task needs more detail. WBTDL task titles are limited to 200 characters, so a task isn't the place for a stack trace or a complete specification.

Leave an accurate handoff

At the end of a session, ask the agent to compare what it actually completed with the task. If review is part of that task, leave it open until review is finished. When the agreed work is complete, ask the agent to mark “Show a clear error when a card payment fails” done.

For unfinished work, save a concrete next action such as “Check payment error copy on mobile.” Have the agent look for an existing task before creating a new one. Put the investigation details in the repository or issue so the next session can find them.

What survives a new session?

Saved tasks and their statuses can be fetched again. Chat history, unsaved files, and the agent's assumptions aren't stored by this connection. Start the next session with a fresh read, even if you use the same client. If you switch clients, the guide to one list across multiple assistants explains the handoff.

MCP task access also has retrieval and editing limits. It doesn't make the coding agent run in the background or turn a task marked done into proof that the code works. Check the change before closing the task.