Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Explore what Zapier MCP can do for the user — interview them about their role and the apps they live in, suggest specific use cases as on-demand prompts, then walk them through enabling the actions to make those use cases real. The natural next step after `zapier-demo`. Use when the user asks "what else can Zapier do for me", "set up more tools", "add a starter pack for my role", "what should I enable next", "suggest workflows", "help me figure out what to do with Zapier", "I don't know where to
.claude/skills/zapier-zapier-explore/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | 202% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 138% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 113% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 197% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 85% | 0% |
Help the user expand beyond the first action — figure out what Zapier MCP can do for them based on their role and apps, then walk them through enabling a starter set of actions tailored to that work.
This is the natural follow-on to zapier-demo. Demo proves one action works; explore turns that into a real toolkit for the user's day-to-day.
For how the Zapier MCP server itself works, see docs.zapier.com/mcp.
If the user hasn't run their first action yet, route to zapier-demo first — explore works best after a win.
Keep it short — two or three questions, not a survey. Adapt follow-ups based on answers.
Open with:
> "Let's figure out what to add to your Zapier toolkit. Quick context first — what do you do for work, and what apps are you in every day?"
Listen for:
If they give you only role or only apps, ask one follow-up. If they give you both, optionally ask:
> "Is there a task you keep redoing manually that you'd love to delegate?"
Don't push past three questions. Move on with what you've got.
Pick the role library below that best matches the user's answers (multiple is fine — combine for hybrid roles). For each, surface 4–6 use cases tailored to the apps they mentioned. If they named an app you don't have in the library, default to the closest equivalent (e.g., "Outlook" → use the Gmail patterns; "Monday.com" → use the Asana patterns).
For each use case, output:
Format as a short prose block per use case, not a table. Don't list more than 6 use cases at once. If you have more, present them in waves — give 4, see if any land, then offer "want more?"
For the use cases the user reacts to ("yes that one"), produce a consolidated enable list grouped by app:
> "To make those work, you'll need to enable: > - Slack: Send Channel Message, Find Message > - Jira: Find Issue by Key, Create Issue > - Google Calendar: Find Events"
Cross-reference the Recommended actions by app table below to fill any gaps — aim for 2–4 actions per app (1–2 search, 1–2 write).
Direct the user to their Zapier dashboard:
get_configuration_url tool, call it first and give them the direct link.Then tell them what to do:
> "Open that link], find your server, and add the actions in the list above. You'll also need to connect each app's account when prompted (OAuth). Come back and say done when everything is added."
Wait for confirmation. If they hit issues:
Re-inspect the available Zapier MCP tools and confirm the new actions are present. If anything is missing, troubleshoot with the user — most often a client reload is enough.
Once everything is enabled, name the win and offer a natural next step:
> "You're set up with N] tools across App list]. Try one now — say something like 'example prompt tailored to their setup]' and I'll run it. Or run /zapier-status anytime to check the health of your tools."
Don't read the user every use case verbatim. Pick the most relevant ones for their context and present them as natural recommendations.
Use this as the reference when building the enable list in Step 3. Aim for 2–4 actions per app — one or two search actions and one or two write actions.
| App | Search actions | Write actions | | --------------- | -------------------------------------- | ------------------------ | | Slack | Find Message, Get Message | Send Channel Message | | Gmail | Find Email | Send Email, Create Draft | | Google Calendar | Find Events | Create Detailed Event | | Google Docs | Get Document Content | Create Document | | Google Sheets | Get Data Range, Lookup Row | Add Row | | Jira | Find Issue by Key, Find Issues via JQL | Create Issue | | Linear | Find Issue | Create Issue | | GitLab | Find Merge Requests | (read-heavy by nature) | | GitHub | Find Issue, Find Pull Request | Create Issue | | HubSpot | Find Contact, Find Company | Create Contact | | Notion | Find Page, Find Database Item | Create Page | | Zoom | Find Meeting | (read-heavy) | | Coda | Find Row | Create Row | | Airtable | Find Record | Create Record |
discover_zapier_actions is exposed by the server): if the user names something you're unsure about, you can silently call discover_zapier_actions to verify before recommending. Don't surface that detail unless the user asks.When presenting a use case, use this shape — short prose, not a table row:
> What the use case does, in plain language] > > Say to me: "[exact prompt in quotes]". I'll what the action does in one line]. > > Actions to enable: App]: Action], Action]
Concrete, never abstract. Avoid "Zapier can help you streamline your workflow." Say instead "Try saying to me: 'Find the latest 3 Slack messages from #product-feedback' and I'll pull them for you." Show, don't pitch.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-20 | pass→pass | 14,355 | 6,664 | -54% | 1 | 1 | 0% | 2,307 | 4,939 | +114% | 0 | 0 | — |
case-05 | pass→pass | 6,701 | 3,918 | -42% | 1 | 1 | 0% | 1,013 | 4,494 | +344% | 0 | 0 | — |
case-15 | fail→pass | 9,301 | 3,887 | -58% | 1 | 1 | 0% | 1,489 | 4,496 | +202% | 0 | 0 | — |
case-01 | fail→fail | 6,528 | 26,966 | +313% | 1 | 1 | 0% | 1,054 | 4,158 | +294% | 0 | 0 | — |
case-02 | fail→pass | 13,081 | 8,191 | -37% | 1 | 1 | 0% | 2,225 | 5,302 | +138% | 0 | 0 | — |
case-03 | fail→pass | 15,009 | 10,695 | -29% | 1 | 1 | 0% | 2,542 | 5,421 | +113% | 0 | 0 | — |
case-04 | pass→pass | 13,690 | 8,377 | -39% | 1 | 1 | 0% | 2,311 | 5,239 | +127% | 0 | 0 | — |
case-06 | pass→pass | 7,945 | 6,669 | -16% | 1 | 1 | 0% | 1,306 | 4,843 | +271% | 0 | 0 | — |
case-07 | fail→pass | 8,655 | 1,684 | -81% | 1 | 1 | 0% | 1,363 | 4,043 | +197% | 0 | 0 | — |
case-08 | fail→pass | 16,505 | 6,259 | -62% | 1 | 1 | 0% | 2,588 | 4,797 | +85% | 0 | 0 | — |
case-09 | pass→pass | 14,365 | 5,614 | -61% | 1 | 1 | 0% | 2,197 | 4,708 | +114% | 0 | 0 | — |
case-10 | fail→pass | 14,229 | 7,233 | -49% | 1 | 1 | 0% | 2,314 | 5,129 | +122% | 0 | 0 | — |
case-11 | pass→pass | 14,862 | 6,076 | -59% | 1 | 1 | 0% | 2,344 | 4,797 | +105% | 0 | 0 | — |
case-12 | pass→pass | 22,466 | 7,297 | -68% | 1 | 1 | 0% | 2,359 | 5,065 | +115% | 0 | 0 | — |
case-13 | pass→pass | 13,922 | 8,785 | -37% | 1 | 1 | 0% | 2,456 | 5,341 | +117% | 0 | 0 | — |
case-14 | fail→pass | 16,121 | 5,095 | -68% | 1 | 1 | 0% | 2,525 | 4,652 | +84% | 0 | 0 | — |
case-16 | pass→pass | 9,507 | 3,405 | -64% | 1 | 1 | 0% | 1,595 | 4,330 | +171% | 0 | 0 | — |
case-17 | fail→pass | 8,227 | 3,569 | -57% | 1 | 1 | 0% | 1,438 | 4,428 | +208% | 0 | 0 | — |
case-18 | fail→fail | 3,247 | 3,547 | +9% | 1 | 1 | 0% | 452 | 4,361 | +865% | 0 | 0 | — |
case-19 | pass→pass | 10,204 | 5,078 | -50% | 1 | 1 | 0% | 1,789 | 4,643 | +160% | 0 | 0 | — |
case-21 | pass→pass | 13,497 | 2,516 | -81% | 1 | 1 | 0% | 2,283 | 4,188 | +83% | 0 | 0 | — |
case-22 | pass→pass | 8,316 | 4,664 | -44% | 1 | 1 | 0% | 1,390 | 4,648 | +234% | 0 | 0 | — |
case-23 | pass→pass | 4,712 | 4,355 | -8% | 1 | 1 | 0% | 715 | 4,496 | +529% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +35 percentage points is the difference between those two pass rates over the 23 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.