Documentation menu

Tool reference

Every tool a connected assistant can call, grouped the same way the server registers them. A tool is only ever offered to the assistant if your connection holds the scope it needs — there's no such thing as a write tool that's present but always errors; if you didn't grant contacts:write, the assistant simply doesn't see upsert_contacts at all. See Scopes & permissions for what each scope means.

Destructive in the tables below is the same MCP annotation Claude's own review process requires — it marks a tool whose effect can't be trivially undone (deleting a project, overwriting a tree, removing a contact), not "dangerous" in some vaguer sense. Every write tool still only takes the fields you'd expect; none of them accept a raw payload the model could smuggle arbitrary changes through.

Discovery

ToolScopeNotes
searchany of projects:read / templates:read / contacts:readSearches projects, templates and contact lists by name. Returns opaque ids for fetch. This is the tool a deep-research-capable client calls automatically.
fetchsame as searchResolves one id from search into its full content.

Email design

ToolScopeDestructiveNotes
list_projectsprojects:readId, name, purpose for every project.
get_emailprojects:readA project's preheader, current block tree, theme and response settings.
preview_emailprojects:readRenders through the real mailed-HTML pipeline — the exact document a send would produce. Widget-attached (see below).
list_block_typesprojects:readThe full 44-block catalog, generated from the live registry — always current, never hand-maintained.
list_templatesprojects:readYour saved templates, or the approved community gallery.
create_projectprojects:writeOptionally seeded from a built-in purpose template, with an optional preheader.
add_block / update_block / move_blockprojects:writeTree edits, canPlace/field-validated the same way Mia AI's edits are.
remove_blockprojects:write✓Deletes a block and its children.
set_themeprojects:writeBrand colors, font, type scale, button styles, link style, block spacing, card layout and dark mode.
rename_projectprojects:write
delete_projectprojects:write✓Permanent.
apply_templateprojects:write✓Overwrites the project's entire tree and theme.
save_as_templatetemplates:writeSnapshots the current project into a reusable template.
publish_landing_pageprojects:writePublishes an email as a public web page at /lp/<slug> and returns its URL. The assistant is told to show you the email and get your approval first, as with sending.
unpublish_landing_pageprojects:writeTakes the page down. The address stays reserved for you.
set_response_settingsprojects:writeSets which answers make a complete response: every counted block or any answer, which blocks count, the final message and an optional https redirect.

Audience

ToolScopeDestructiveNotes
list_data_sourcescontacts:readEvery data source, contact lists included. Credential values are never returned.
list_contact_listscontacts:readContact lists specifically — id, name, fields, row count.
get_contact_list_summarycontacts:readCounts, suppression totals and the email verification breakdown, plus the last re-check's status — no individual contact data.
find_contactscontacts:readBounded lookup (≤25 rows, cursor-paginated). sortBy: "engagementScore" returns the list's warmest contacts first, with their score and this week's change. Every call is written to the personal-data access log, same as a dashboard contact read.
upsert_contactscontacts:writeAdd or update rows; enforces your plan's contact cap.
remove_contactcontacts:write✓Permanently removes one row.
add_suppression / remove_suppressioncontacts:writeBlocks or unblocks an address account-wide, same effect as a recipient's own unsubscribe link.

Results

ToolScopeDestructiveNotes
get_campaign_resultsanalytics:readOpens/clicks/responses/revenue, A/B variant comparison, click heatmap. Summarized only — never raw per-recipient events. Widget-attached.
get_email_insightsanalytics:readOne Insights view (overview, blocks, one block, audience, crosstab, sends or engagement) as aggregates, under optional date, send, variant and list filters. The same numbers as the page; audience and crosstab need Pro or above. Differences are observed, not tested.
get_form_responsesanalytics:readPer-question rollups; free-text answers are a capped, fenced sample.
list_ordersanalytics:readProduct-block checkout orders.
list_campaign_sendsanalytics:readPast bulk sends, most recent first.
list_schedulesanalytics:readRecurring sends configured on a project.
mark_order_fulfilledcommerce:writeMarks a physical order shipped and emails the buyer.
create_schedulecampaigns:sendConfigures a new recurring send.
set_schedule_enabledcampaigns:send✓Marked destructive because disabling a live schedule stops future sends silently.

Sending

ToolScopeDestructiveNotes
send_test_emailcampaigns:readA single [Test]-prefixed copy to one address you specify — doesn't touch your real audience or need campaigns:send.
prepare_sendcampaigns:sendRead-only by annotation — resolves the audience and renders a preview, sends nothing. Returns a confirmationToken, and how many contacts will be skipped as invalid addresses or for no consent. Widget-attached (send-confirmation card).
confirm_sendcampaigns:send✓Requires the exact confirmationToken prepare_send returned, unchanged, within 10 minutes. See Security for why.

Mailbox & tickets

ToolScopeNotes
list_mailbox_threads / get_mailbox_threadmailbox:readInbound mail at your verified domain. Message bodies are fenced as untrusted, recipient-authored content.
reply_to_mailbox_threadmailbox:writeSends from the exact mailbox address the original mail arrived at, correctly threaded.
list_tickets / get_ticketmailbox:readSupport-ticket conversations. Same untrusted-content fencing.
reply_to_ticketmailbox:writeAdds an owner message and emails the recipient a link back.

Journeys & segments

ToolScopeDestructiveNotes
list_journeysjourneys:readId, name, on/off, trigger kind and step count for every journey.
get_journeyjourneys:readA journey's full flow, with your emails', lists' and segments' names, plus anything that would stop it from running. For a journey that's off and can run, also returns an enableToken.
list_segmentsjourneys:readSaved segments, with their contact list and rules.
create_segmentjourneys:writeSaves a segment on one contact list. Enforces your plan's segment cap.
build_journeyjourneys:writeBuilds a new journey, or edits one that's off, from a plain-language request, using the journey builder's Mia AI. Uses one Mia AI turn. The result is always saved turned off, and a journey that's on is never changed.
set_journey_enabledjourneys:write (+ campaigns:send to turn on)✓Turning a journey on requires the enableToken from get_journey, unchanged, within 10 minutes and before any further edit. Same two-step idea as prepare_send/confirm_send. Pausing needs neither.

Experiments

ToolScopeDestructiveNotes
list_experimentsanalytics:readYour A/B experiments, newest first, both send and journey ones: status, arms, metrics, the analysis engine's latest result, any recorded decision and a link to the results page. Filters by status, kind, email or journey.
get_experiment_resultsanalytics:readOne experiment's report, the same as the results page's JSON export, plus the engine's outcome as one line. The assistant is told that line is the only verdict, so it names a winner only when the engine did.
create_experiment_draftprojects:writeCreates a draft send experiment on one email: name, hypothesis, metrics, and up to your plan's number of arms, split evenly. It sends nothing. You set the audience, lock and launch it in the dashboard, from the link it returns. Journey experiments are created from a split step instead.

No tool locks, launches, decides or deploys an experiment. Those stay your own steps in the dashboard.

Account

ToolScopeNotes
get_plan_usagenone — always availablePlan, quotas, and this month's usage. Read-only, and not sensitive account data the way projects/contacts/mailbox are, so it's available regardless of which scopes you granted.

Widgets

Three tools attach an interactive widget on hosts that support MCP Apps (Claude — Gemini has no widget layer and always gets the plain-text/JSON result instead): preview_email (live desktop/mobile email preview), get_campaign_results (results dashboard), and prepare_send (the send-confirmation card, whose own Confirm & Send button is what actually calls confirm_send).