MCP server
Prompt templates
Seventeen read-only prompt templates. Fifteen name the tools an agent should
call, in order, to produce one deliverable, and take workspace (a name or
id — resolve it with list_workspaces first) plus their own arguments. The
last two (local_pitch, national_pitch, local/national AEO Task 11, spec
§13) are different in shape: they take an already-resolved workspace_id
directly (not a name to resolve) and render fixed agency-pitch workflow
text rather than a tool-calling sequence — the assistant reading one is
told to read the workspace's own stored, confirmed market and report
snapshot itself (get_market_profile, get_client_report), never to
change it, start collection, or send anything. Every template embeds no
tenant data, key or host, and is visible to any authenticated principal
through prompts/list; the tools a tool-calling template names are still
gated by that principal's own tools/list, so a template can be read by an
agent that cannot execute every step it names — the agent should say so
rather than guess.
| Template | Arguments | Tools it calls |
|---|---|---|
weekly_visibility_brief | workspace, week? (ISO week like 2026-W36; default the last complete week) | get_overview, get_report(kind=brands, dimensions=[engine]), list_prompts |
daily_watchlist_triage | workspace, date? (YYYY-MM-DD; default yesterday) | search_answers, get_answer, get_report(dimensions=[date]) |
citation_gap_analysis | workspace, start_date, end_date | get_report(kind=domains) with gap + source_bucket, list_urls, get_source_content |
competitor_change_report | workspace, competitor, start_date?, end_date? (default the last 30 days) | get_report(kind=brands, dimensions=[week]), search_answers(brand_ids) |
content_opportunity_plan | workspace, start_date?, end_date? (default the last 30 days) | get_report(kind=urls), list_source_classifications, get_source_content |
claim_accuracy_audit | workspace, topic? (topic slug from list_topics) | search_answers, get_answer, get_brand_profile |
client_executive_summary | workspace, period ('last_30_days', 'last_quarter', or 'YYYY-MM-DD..YYYY-MM-DD') | get_overview, get_report(kind=brands), get_answers_report |
agency_portfolio_review | workspace (agency org; no arguments beyond it) | list_workspaces, get_portfolio_summary, get_overview per workspace |
new_business_visibility_audit | workspace, prospect_domain | get_brand_profile, list_prompt_candidates, get_report(kind=domains) |
engine_scorecard | workspace, start_date?, end_date? (default the last 30 days) | list_engines, get_report(kind=brands, dimensions=[engine]), get_answers_report(dimensions=[engine]) |
topic_heatmap | workspace, start_date?, end_date? (default the last 30 days) | list_topics, get_report(kind=brands, dimensions=[topic_id]) |
prompt_set_grader | workspace, start_date?, end_date? (default the last 30 days) | list_prompts, get_report(kind=brands, dimensions=[prompt_id]), get_answers_report(dimensions=[prompt_id]) |
campaign_impact | workspace, start_date, end_date | get_report(kind=brands, dimensions=[week], compare_to), get_report(kind=domains, dimensions=[week]) |
first_session | workspace? (ignored; walks every workspace the token can see) | get_account_context, list_workspaces, get_onboarding_status, validate_brand_setup per workspace, get_client_report(view=audit, gaps, actions) |
setup_review | workspace | get_onboarding_status, get_market_profile, list_topics, list_prompts, validate_brand_setup; writes (update_prompts, archive_prompts, import_prompts, create_topics, add_competitors, dry_run first) only with the user's consent and only if this connection's role grants them, else points to the studio |
local_pitch | workspace_id, audit_id, business_name, city_name?, region_name? | none named — pitch workflow text; the assistant reads get_market_profile/get_client_report itself |
national_pitch | workspace_id, audit_id, business_name, city_name?, region_name? (unused) | none named — pitch workflow text; the assistant reads get_market_profile/get_client_report itself |
prospect_pitch_deck | prospect, format? (pptx default, or pdf) | list_prospect_reports, then get_prospect_deck — hands over the exact file the app downloads (D10); only when the user asks for changes: get_prospect_report, get_brand_kit, get_brand_asset, get_agency_brand_kit, same 10 slides, the kit's colours and safe-list fonts only. Listed only while prospect reports are enabled |
first_session is the sequence the server's own instructions recommend
for a newly connected assistant.
Fetch the list with prompts/list; render one with prompts/get, passing
its arguments as plain strings (dates as YYYY-MM-DD, workspace as the
name or id list_workspaces returned). The rendered text is a set of
numbered instructions for an agent — it is not itself a report, and no
template argument accepts anything but a string.