Tools and prompts
The server lists fifteen tools. Most work happens in a few of them; everything else in the dashboard is reached through find_actions and the four run_* tools.
Almost every tool takes a site_id. Agents get it from list_sites, which is why it comes first. Periods are today, yesterday, 7d, 30d, 90d, 180d, 365d, month, lastmonth, ytd, all, or a range like 2026-10-01..2026-10-31. Money is in US dollars.
Read
| Tool | What it does |
|---|---|
list_sites | Every site you can open, grouped by workspace, with your role, whether it is live, its plan or lock, and today's numbers. |
get_overview | One site's headline numbers against the previous period, the daily trend and the top sources, pages, countries, devices and campaigns. |
query_analytics | Any metric by any dimension, with filters. Use it for questions the overview doesn't answer. |
explain_change | What drove a rise or drop between two periods, by source, page, country, device and campaign. |
get_errors | Issues ranked by people hit and revenue at risk. With issue_id it returns the full detail an agent needs to fix it: stack with the code around it, breadcrumbs, the typical visit, linked feedback and nearby recordings. |
get_feedback | Visitor notes. With feedback_id it returns the text, console errors and failed requests, screenshots as images, the linked issue and the recording. |
get_sessions | Find recordings, or read one session as a text log around a moment. |
check_installation | Whether the script loaded, the first visit arrived and each feature is reporting, and what is missing. |
verify_fix | The production check for a fix you marked. See Fix errors with an agent. |
Set up
| Tool | What it does |
|---|---|
set_up_site | Adds a website and returns install steps for its framework, with the features that fit turned on. |
Everything else
| Tool | What it does |
|---|---|
find_actions | Searches the dashboard's other actions and shows the exact arguments each one takes. |
run_read | Runs a read action. Changes nothing. |
run_create | Runs an action that adds something: a goal, funnel, alert rule, chart note, tag or draft reply. |
run_update | Runs an action that changes something and is safe to repeat: resolving issues, a note's state, settings. |
run_delete | Runs an action that removes something. Clients ask you first. |
Splitting them by kind lets a client treat each differently: reads run freely, deletes ask. With Read only access the three writing tools refuse.
Discover an action
Ask the agent for what you want in plain words, such as "add a signup goal" or "list alert rules". It calls find_actions with a few words, gets back the matches with their run_* tool and arguments, and runs one. Names look like area.verb, for example goals.replace, funnels.create, notes.list and issues.update.
find_actions also lists what an agent can't do and where to go in the dashboard instead. See Security.
Prompts
Prompts appear as slash commands in clients that show them. Each one starts a whole piece of work.
| Prompt | What it asks the agent to do |
|---|---|
fix_issue | Read one error issue, fix its cause in your code, ship it and mark it fixed. |
fix_feedback | Fix what a visitor reported, then set the note to done. |
fix_top_errors | Take the three costliest errors that come from your code and fix each. |
set_up_analytics | Install SnowAnalytics, wait for the first visit and create goals for the signups or purchases it finds. |
weekly_report | A short report on the last week: traffic, what moved, errors and feedback. |
why_did_it_change | Explain a rise or drop in visitors or revenue. |
review_feedback | Group open feedback into themes and say what to fix first. |