Choose the first Claude connectors to try

Help me catch the replies I owe. Review Gmail threads from [sender or domain] between [dates and time zone] about [project]. If Gmail is unavailable, use the thread excerpts I paste. For each request, check whether I answered it later in the same thread before calling it open.

Give me a table with who asked, what they need, the stated deadline, and a link or subject/date I can verify. Separate genuine follow-ups from FYIs and already answered requests. Draft one reply for the most urgent open item. Leave prices, promises, and dates I haven’t approved in brackets. Keep the draft in this chat.

Help me get ready for [date] using [calendar name] in [time zone]. If you can’t see it, use [pasted schedule]. Include only [working hours] and keep personal event details out of the brief.

Give me a time-ordered agenda with organizer, event link, and one prep action based on the event description. Call out overlaps and meetings without an agenda. Separate prep I can do now from information I need to ask for. End with the one meeting I should prepare for first and why. Keep calendar events and invitations unchanged.

Find the version of [document or deliverable] I should work from in [folder or project]. I know these clues: [owner, approximate date, phrase, or prior link]. If you can’t read Drive, ask for a file list or pasted candidate links.

Return up to five matches with title, folder, owner if visible, modified date, and source link. Explain which version looks authoritative based on naming, location, or contents. A newer modified date alone doesn’t settle it. If two versions look current, give me the difference I should check before choosing. Keep files and permissions unchanged.

Find the current instructions for [process] inside [workspace, database, or page]. Use only the pages you can open. If Notion is unavailable, ask me for the relevant page text or export.

Return a short “do this today” checklist with a source link beside each important action. If two pages disagree, show the two instructions, their dates or owners if visible, and the decision I need to make. Flag missing steps separately. Draft one proposed edit in chat if it would clear up the biggest ambiguity. Leave the workspace untouched.

Help me reconstruct what we decided about [topic]. Read [channel or thread] for [date range]. If you can’t open it, use the thread excerpts I paste. Follow replies so you can see when an early idea was changed or dropped.

Give me decisions with links to the messages that settle them, open questions, and next actions with an owner or due date only when someone explicitly named one. Mark implied actions as suggestions. End with a short handoff I could paste to a teammate after checking it myself. Keep Slack unchanged.

Turn the transcript for [meeting and date] into a handoff for [person or team]. If you can’t access it, ask me for an approved export. Use speaker names as written and flag any uncertain transcription.

Give me the decisions, commitments, and unresolved questions. For each commitment, include the owner, due date if explicitly stated, and a timestamp or short source excerpt. Separate “we might” from “I’ll do it.” Then draft a short handoff with the three things the reader needs to know. Keep it in this chat for review.

Turn [recording transcript or export] into a help article for [reader] who wants to [task]. If all you have is audio text, mark clicks and UI details that the transcript doesn’t establish.

Start with what someone should have open. Write the steps using button or menu names only where the recording confirms them. Add a quick way to tell whether each tricky step worked and a short “if you get stuck” section. Give me a screenshot list with timestamps and captions, where available. Flag any step I need to verify in the video before publishing.

Help me turn one change to [page or project brief] into a request I can paste into Lovable. The visitor is [audience] and needs to [goal]. Here’s the current screenshot or copy: [attachment]. If you can’t see the page, work from what I attach and tell me what you couldn’t check.

Find the first point where someone may hesitate. Return the current wording or layout, the proposed change, and three checks for phone width, keyboard use, and the main action. Name which sections and behaviors should stay in place. Write the final Lovable prompt in a separate block for me to review.

Read responses for [form name] from [start] to [end] or use [redacted export]. I’m trying to decide [question]. Skip personal contact information and show any excluded or missing rows in the totals.

Group repeated requests and give me the count and denominator for each theme. A response may fit more than one theme, so say how you counted. Show one anonymized example per major theme and the question people most often skip. End with one form wording change to test, with old and new copy. Keep the form and responses unchanged.

Check the listings for [items or collection] in [store], or use [product export]. I’m checking whether the collection is ready for [launch or refresh].

Give me a table with product title, listing status, collection, link or ID, and anything that looks inconsistent. Separate duplicate candidates from confirmed duplicates. If a price or variant differs, show both source records so I can decide which is right. End with three checks I can do as a shopper, including whether the right variant is available. Keep products, pricing, discounts, and orders unchanged.

Help me reconcile this [redacted export or approved payment report] for [date range], in [currency]. Use only the records I provide or authorize. Tell me which date and status fields you used and flag pending or failed items separately.

Give me total successful payments, refunds, and net receipts, with the calculation and source rows. Break out fees if they’re present and say whether your net figure includes them. Keep mixed currencies separate. Flag duplicate IDs, unmatched refunds, and gaps without assuming why they happened. Give me two totals I should compare with the original report. Keep this read-only.

Compare [campaigns] for [date range] using [report or redacted export]. I want to learn [question]. Tell me whether the same attribution window, audience rules, and metric definitions apply to both campaigns before comparing conversion.

Return a table with delivered count, unique clicks, click rate and denominator, and the conversion metric both reports support. Mark fields you couldn’t verify. Suggest one variable to test next and the metric that would tell me whether it helped. Keep campaigns, segments, and flows unchanged.

Review [pipeline] for records in [stage] that haven’t had a recorded reply or owner activity since [date]. Use a redacted export if you can’t open the pipeline. Exclude closed and opted-out contacts, and explain how you decided a record was stale.

Return counts by age and a short table with record ID, last relevant activity, and the reason to follow up or wait. Flag any record whose activity is unclear. Draft two neutral follow-ups I can review, with unverified details in brackets. Keep contacts, messages, pipeline stages, and automations unchanged.

Debug this Zap from [redacted test log], [expected behavior], and [the steps that ran]. Find the first step where the output differs from what the next step needs. If the log doesn’t include its input or output, tell me exactly what to copy from the test history.

Give me a compact table with step, observed value, expected value, and a likely cause. Propose one change to test in a safe copy. Include a small example payload and the expected output, plus a second test for a missing value. Keep the live Zap and its runs untouched.

Review [dataset or export] to help answer [research question]. Use the data I supply. Start by checking whether its columns, dates, and coverage are enough for that question.

Give me the raw row count, the duplicate rule you used, the count after deduplication, and the fields most often missing. For each pattern, give its count, denominator, and a row or URL I can verify. Separate what the dataset supports from what would need a new sample. End with one practical next question and a simple check I can do on the existing records. Keep collection runs unchanged.