Bring your thrw lists into Claude Code and Codex.
Connect through MCP so the AI you already use can read and update your thrw lists.
Connecting another AI to thrw lets you continue with the same lists and notes instead of pasting a plan into every chat. The handoff works best when you verify both reading and a small change.
Connect your account
In thrw's settings, open the integrations area and choose the guide for your AI client. The MCP server address is:
https://api.thrw.cc/mcp
Add that remote server in Claude Code or Codex using the relevant connection guide. Complete the thrw sign-in and account authorization when prompted. A configured URL alone doesn't establish access to your account.
Use the instructions for your client; the desktop app, CLI, and editor integrations have different setup screens. If you have more than one account, confirm that you authorize the one containing your plan.
The in-app guide also offers an installation request for the THRW planning skill. Use it if you want the client to recognize THRW planning workflows, and verify the skill separately. Account access and skill discovery are two different checks.
Verify reading before asking for changes
Start with a list you can recognize:
Read my presentation list in thrw, including descriptions and completion states. Don't change anything yet.
Look for something specific, such as the draft deadline or the note about waiting for sales figures. A general reply about how to give a presentation doesn't prove that the connection is working.
Make one small, visible change
Add “Check slides 1–8 for typos” to that presentation list. Leave it undated and keep the existing items.
Review any approval requested by your AI client. Open thrw and check that one new item appeared in the correct list. If it isn't there, have the AI read the list again before asking it to create the same item a second time.
Leave the next AI enough to continue
After working, ask the first AI to record the outcome:
I checked slides 1–8 for typos. Mark the typo check complete and add that scope to its description. I haven't checked the speaker notes.
In a second connected client, ask:
What was checked in the presentation, and what still needs attention? Read the recorded notes.
The second AI should be able to distinguish checked slides from unchecked speaker notes. This shares the recorded plan and progress, not the first client's entire private chat history. Ask it to write down decisions or findings you want the next client to use.

