How to Make a Claude Code Routine Publish a Weekly Report to the Same URL

The short answer

Give the routine one job. Build the report. Then publish it to the Display artifact that already exists, so the link does not change.

A routine is a saved Claude Code configuration. It holds a prompt, repositories, and connectors. It runs on a schedule without your laptop. Connect Display to the routine, and end the prompt with an instruction to update the existing artifact instead of creating a new one. Set access to Company, so colleagues open it with their Google or Microsoft account or a one-time password (OTP).

Checked on 14 October 2026 against the Claude Code documentation.

What a routine is

Anthropic's documentation says a routine is a saved Claude Code configuration: a prompt, one or more repositories, and a set of connectors. It runs on Anthropic-managed cloud infrastructure, so it keeps working when your laptop is closed. Routines are in research preview, so details may change.

Facts that matter for a weekly report:

• Routines are available on Pro, Max, Team, and Enterprise plans.
• You create one at claude.ai/code/routines, in the Desktop app, or with /schedule in the CLI.
• A schedule trigger offers hourly, daily, weekday, and weekly presets. The minimum interval is one hour.
• A run that you schedule exactly on the hour can start several minutes late. Pick a minute past the hour, such as 9:07.
• A routine belongs to your claude.ai account. Teammates do not share it, and its actions appear as you.
• A routine runs without permission prompts. It can use every tool in the connectors you include.

Anthropic describes two other ways to schedule work. Desktop scheduled tasks run on your machine. /loop runs inside an open session, and a recurring /loop task expires after seven days. For a weekly report that must run without your machine, use a routine.

Step 1: publish the first version yourself

Do this once, in a normal session. Ask your agent:

Publish this report to Display as a Markdown artifact named "Weekly KPI report".
Set access to Company. Tell me the link.

You now have one artifact with its own link. Open it and check the access. Send the link to your readers.

Step 2: connect Display to the routine

Routines use connectors, which are the MCP integrations on your claude.ai account. Anthropic's documentation says MCP servers that you added locally with claude mcp add do not appear in that list. Add Display as a connector at claude.ai/customize/connectors, or declare it in a committed .mcp.json file in the routine's repository.

Remove connectors the routine does not need. Anthropic notes that Claude can use every tool in an included connector, including writes, without asking.

Check the access path before the first scheduled run. The routine publishes with the connector, so it acts for the person who connected it. Display records the agent and that person on each version.

Step 3: write the routine prompt

The prompt must be self-contained. The routine has no one to ask. Use this as a start and replace the data steps with yours.

You write the weekly KPI report for the product team.
 
1. Read the metrics from the sources listed below. Use only data from the last 7 days.
2. Write the report in Markdown. Use these headings: Summary, Numbers, What changed, Risks.
   Put the week's date range under the title.
3. Find the Display artifact named "Weekly KPI report". Do not create a new artifact.
4. Read the open comments on that artifact. If a reader asked a question, answer it
   in the new version. Reply to the comment and say what changed.
   Resolve a comment only if you fully answered it.
5. Publish the new report as an update to that artifact. Keep the same access.
6. If the update fails because someone else edited the artifact first, read the
   newer version, apply your changes to it, and publish again.
7. Reply with the artifact link and a two-line summary.
 
Sources: <list your sources here>

Why each step is there:

• Step 3 keeps the link the same. Without it, the agent may create a new artifact each week.
• Step 4 closes the loop. Comments stay on the version where the reader made them, so last week's questions do not disappear.
• Step 6 handles a person who edited the report in the browser between runs.

Step 4: set the schedule

At claude.ai/code/routines, select New routine. Paste the prompt. Choose the weekly preset and a time. Review the connectors. Create the routine, then select Run now once to test it.

A green status in the run list means the session started and ended without an infrastructure error. It does not mean the report is right. Open the run and read what the agent did.

What your readers see

Readers open the link they already have. They sign in with their company account. They see the newest version. They select a passage and comment. If they have edit access, they can fix a number in the page.

Each run is a stored version. Display shows who is viewing and which version each person looks at. Published versions and comments are attributed to the agent and the person it acted for.

Outside guests can open an artifact when you invite them, but they cannot comment yet.

Three more routine examples

These follow the same pattern: one artifact, updated on a schedule.

Release notes. A GitHub trigger runs when a release is published. The prompt reads the merged pull requests, writes the notes, and updates the artifact named "Release notes".

Competitor watch. A weekly schedule. The prompt reads the competitor pages you list, notes what changed, and updates "Competitor watch". Readers comment on items that need a decision.

Incident review. An API trigger fires after an incident closes. The prompt gathers the timeline you pass in, drafts the review, and publishes a new artifact for each incident. This one creates a new link on purpose.