Track Claude on your own Claude plan

How it works

You already pay for Claude. With this setup, a scheduled task in your own Claude account asks Claude your tracked prompts once a day, the same way a person would ask them in Claude with web search on. The task sends Claude's answers to Spyglasses, and they show up in your dashboards next to ChatGPT, Gemini and Google AI Overviews.

• Your Claude login stays with you. Spyglasses never asks for, sees or stores your Claude password, login or session. The task runs in your Claude account and sends only the answers.
• It runs on its own. Once it is set up, the task runs in Claude's cloud every day. Your computer can be off.
• You choose the prompts. Turn Claude on for a project in Spyglasses and that project's prompts are added to the next day's run.

One thing to know before you start: no locations

Claude's web search doesn't take a location. Answers from this setup are not specific to a country, market or city, and Spyglasses doesn't run your location or persona versions of a prompt on Claude with this setup. If your results depend on where the person asking is (local services, regional brands, country-specific pricing), talk to us about other ways to track Claude.

What you need

• A Claude Pro, Max, Team or Enterprise plan, signed in at claude.ai.
• Your brand's Spyglasses API key. We send it with your setup message, and you'll also find it in your property settings. It's the same key your website plugin uses.
• About 15 minutes.

How much of your Claude plan this uses

Claude plans have a five-hour limit and a weekly limit, shared by everything you do in Claude. Each tracked prompt uses a small share of both. These figures come from our own tests on Pro and Max accounts in November 2026. Treat them as a guide: they vary with the prompt, and your own Claude use counts toward the same limits.

Each prompt takes about a minute, so a run of 20 prompts takes about 20 minutes. Schedule runs for hours when you don't usually use Claude, so the tracking doesn't use the five-hour window you need during your working day.

Step 1: Create a Claude environment for Spyglasses

The environment lets the task reach Spyglasses and nothing new beyond that.

1. Go to claude.ai/code, open Environments and choose New environment.
2. Name it Spyglasses tracking.
3. Network access: choose Limited and add this line to Allowed domains: www.spyglasses.io. Don't choose Full. Limited is the setup we tested against answers from claude.ai.
4. Secret: add a network secret for www.spyglasses.io that sends the header x-api-key, with your Spyglasses API key as its value. Claude adds it to requests to Spyglasses on their way out, so the key never appears inside the session.
5. Save.

Step 2: Create the scheduled task

1. Go to claude.ai/code/routines and start a new scheduled task from one of the examples, such as the daily briefing. Starting from an example gives you the cloud environment and custom schedule settings; a task started from a blank form runs as a chat instead.
2. Name: Spyglasses tracking.
3. Model: Opus 5.5, with effort left at Default (the picker under the instructions).
4. Cloud environment: choose Spyglasses tracking (from step 1) in the environment picker beside the model. Repository: none.
5. Trigger: choose Custom and paste the schedule line from your setup message, for example 7 9 * * * for one run a day, or 7 9-11 * * * if your prompts need several runs. Schedules are in UTC. Keep the minutes at 07: runs set on the hour can start several minutes late.
6. Connectors: if the task lists any connectors, remove every one. The task must run with none attached, so your other tools never mix into the answers.
7. Instructions: replace the example's text with the prompt below, exactly.

This is a brand-tracking collection run for Spyglasses. Do not send notifications. Do not use any connector. Do not create or edit files.

Step 1. Run this command and read what it prints:
curl -sS --fail-with-body https://www.spyglasses.io/api/v1/claude-tracking/questions
If it says there are no questions, reply "Nothing to run." and stop.

Step 2. For each question it lists, in the listed order and one at a time, launch one new general-purpose subagent with the Agent tool, using the same model as this session. Give each subagent this instruction and nothing else, with the question text copied exactly:

Use web search to research this question, then answer it the way you would answer the person asking it: <question text>

Do not research or answer the questions yourself, and do not repeat or summarize the answers.

Step 3. After each subagent finishes, and before you start the next question, run this command:
tail -c 600000 "$(ls -t ~/.claude/projects/*/*.jsonl | head -1)" | curl -sS --fail-with-body -X POST -H "Content-Type: application/x-ndjson" --data-binary @- https://www.spyglasses.io/api/v1/claude-tracking/answers

Step 4. When every question is done, reply with the last line the final command printed.

8. Save the task and make sure it is switched on.

Step 3: Test it

1. Open the task and choose Run now.
2. When it finishes, the last line reads something like Received 20 of 20 answers for today.
3. In Spyglasses, open a project with Claude turned on. Latest Claude answers shows the time of the run. Dashboards update within an hour.

Keeping it running

• Adding prompts: turn Claude on for more projects in Spyglasses. The next run picks them up. If you go past what one run holds, ask us for a new hourly line.
• New key: if your brand's API key changes, replace it in the environment's secret before the next run, and in your website plugin.
• Plan limits: if a run reaches your five-hour limit, it stops early. The answers it already sent still count; the rest of that day's prompts show no Claude answer for that day. Lower the number of prompts per run, or move the run to a quieter hour.
• Missing a day: if a run doesn't happen, that day shows a dash for Claude, never a zero, and the next day starts fresh.

Troubleshooting

Questions? Email support@spyglasses.io and include the task's run time.