Capabilities Reference

Capability reference

Generated from packages/core/src/lib/capabilities/catalog.ts. Do not hand-edit — run node scripts/gen-agent-docs.mjs.

Every row is one registry entry; the same entry drives the MCP tool, the /api/v1 request, the public mcp.json, the aicmo run <id> CLI dispatcher, and (where chat is a surface) the AI CMO runtime tool.

Legend — effect: read no state change · internal_write changes AICMOHQ state only · external_write posts/sends outside AICMOHQ · spend consumes credits · delete is irreversible. confirmation: approval = a human must approve before the effect lands, step_up = approval plus re-auth; the kernel enforces external_write ⇒ approval or step_up. * = required input.

Capability id	MCP tool	HTTP (/api/v1)	Scope	Effect	Risk	Confirmation	Surfaces	Inputs	Description
capabilities.list	capabilities_list	— (local)	none	read	low	none	cli, mcp	—	List supported chatbot, CLI, and MCP capabilities with risk and confirmation metadata.
strategy.run	strategy_run	POST /strategy/run	strategy:run	spend	medium	approval	cli, mcp	keywords, platform*	Run the full strategy loop: extract git context and generate ICP-validated drafts into the queue (LinkedIn, X, Reddit).
scout.research	scout	POST /scout	strategy:run	spend	low	none	chat, cli, mcp	keywords	Find marketing opportunities based on product and keywords.
writer.generate	write	POST /write	strategy:run	spend	medium	approval	chat, cli, mcp	platform*, from_git, topic, enqueue	Draft content from a topic or from recent git commits. Set enqueue=true to put the draft in the queue for LinkedIn/X publish.
credits.costs	credit_costs	— (local)	none	read	low	none	cli, mcp	—	List the credit cost of each AICMOHQ agent action. Local, read-only; no auth required.
connections.list	connections_list	GET /connections	connections:read	read	low	none	cli, mcp	—	List connected platforms (check LinkedIn before publishing).
queue.list	queue_list	GET /queue?status={status}	queue:read	read	low	none	cli, mcp	status	List content queue items for the authenticated AICMOHQ account.
queue.get	queue_get	GET /queue/{id}	queue:read	read	low	none	cli, mcp	id*	Fetch one content queue item by id.
queue.delete	queue_delete	DELETE /queue/{id}	queue:write	delete	medium	approval	cli, mcp	id*	Permanently delete a queued draft.
queue.update	queue_update	PATCH /queue/{id}	queue:write	internal_write	low	none	cli, mcp	id*, content*	Update the content of a queued draft.
queue.approve	queue_approve	POST /queue/{id}/approve	queue:write	internal_write	medium	approval	cli, mcp	id*	Approve a draft queue item so it can be published.
queue.publish	queue_publish	POST /queue/{id}/publish	publish	external_write	high	approval	cli, mcp	id*	Publish an approved queue item (requires platform connection, e.g. LinkedIn).
queue.reject	queue_reject	PATCH /queue/{id}	queue:write	internal_write	low	none	cli, mcp	id*	Reject a draft so it will not publish.
queue.best_time	queue_best_time	GET /queue/best-time?platform={platform}&tz={tz}	queue:read	read	low	none	cli, mcp	platform*, tz	Suggest the best publish time for a platform — the workspace’s measured engagement history when available, else the platform’s researched golden-hour window. Advisory only; scheduling still goes through queue_schedule.
queue.schedule	queue_schedule	PATCH /queue/{id}	queue:write	internal_write	medium	approval	cli, mcp	id*, scheduled_for*	Schedule an approved queue item for cron publish (ISO timestamp).
queue.proof	queue_proof	POST /queue/{id}/proof	queue:write	internal_write	low	none	cli, mcp	id*, proofUrl*	Save a live post URL for Hacker News or Product Hunt (no API). Then call queue_publish to mark the item published.
reddit.status	reddit_status	GET /reddit/status	connections:read	read	low	none	cli, mcp	—	Inspect the connected Reddit account’s conservative warm-up stage and engagement limits.
reddit.warmup	reddit_warmup	POST /reddit/warmup	strategy:run	spend	low	none	chat, cli, mcp	keywords, subreddits, limit, productId	Use the connected Reddit OAuth account to find value-first threads. Saves opportunities only; never comments or publishes.
command.chat	command_chat	POST /command/chat	strategy:run	internal_write	low	none	cli, mcp	message*, sessionId, agentId	Send a natural-language mission to AICMOHQ Agent. Persists session memory and returns a proposed plan (approval-first; does not publish). When the response includes a plan, executionPolicy tells you the next step: call command_run with message.id to execute (auto=true only when auto_run_eligible), or command_status with runId to inspect a running one.
command.sessions	command_sessions	GET /command/sessions?limit=20	strategy:run	read	low	none	cli, mcp	limit	List recent AICMOHQ Agent chat sessions (shared memory with dashboard + CLI).
command.run	command_run	POST /command/run	strategy:run	internal_write	medium	approval	cli, mcp	messageId*, auto	Execute a proposed command plan by assistant message id. Rejects auto=true when approval is required. Streams NDJSON progress.
command.status	command_status	GET /command/run/{runId}	strategy:run	read	low	none	cli, mcp	runId*	Inspect a durable command run and its approval-gated steps.
command.approve_step	command_approve_step	POST /command/run/{runId}	strategy:run	internal_write	medium	approval	cli, mcp	runId*, stepId*	Approve one bound CMO step so execution resumes (expires in 15 minutes).
workspace.context	workspace_context	GET /workspace/context	strategy:run	read	low	none	cli, mcp	—	Read the workspace strategy context (product, ICPs, positioning, brand voice) agents ground drafts on.
workspace.get	workspace_get	GET /workspace	strategy:run	read	low	none	cli, mcp	—	Read the raw workspace record (company profile, positioning, ICPs) the agents ground on.
products.list	products_list	GET /products	strategy:run	read	low	none	cli, mcp	—	List the products in the workspace. Most agent work needs a product_id from here first.
products.create	products_create	POST /products	workspace:write	internal_write	low	none	cli, mcp	name*, description, url, target_audience, competitors	Add a product to the workspace so scans and drafts can be scoped to it.
products.update	products_update	PUT /products/{id}	workspace:write	internal_write	low	none	cli, mcp	id*, name, tagline, description, url	Update a product’s name, tagline, description, or url.
products.archive	products_archive	DELETE /products/{id}	workspace:write	delete	medium	approval	cli, mcp	id*	Delete a product. Its past artifacts stay, but new work can no longer target it.
artifacts.list	artifacts_list	GET /artifacts	strategy:run	read	low	none	cli, mcp	—	List agent artifacts (research briefs, drafts, reports) produced for this workspace.
listening.list	listening_list	GET /listening	strategy:run	read	low	none	cli, mcp	—	List captured listening mentions — places the product or its category was discussed.
listening.scan	listening_scan	POST /listening/scan	strategy:run	spend	low	none	cli, mcp	product_id, limit	Run a listening scan for new mentions of the product. Costs credits.
signals.scan	signals_scan	POST /signals/scan	strategy:run	internal_write	low	none	cli, mcp	query*, perSource	Search external sources for demand signals matching a query.
trends.list	trends_list	GET /trends?range={range}	none	read	low	none	cli, mcp	range	Read trending topics for the day or week to ground timely content.
opportunities.list	opportunities_list	GET /opportunities	strategy:run	read	low	none	cli, mcp	—	List scored growth opportunities derived from listening and signals.
opportunities.draft	opportunities_draft	POST /opportunities/{id}/draft	strategy:run	spend	low	none	cli, mcp	id*	Draft content for an opportunity. Costs credits; the draft still needs human approval to publish.
opportunities.queue	opportunities_queue	POST /opportunities/{id}/queue	queue:write	internal_write	low	none	cli, mcp	id*	Move an opportunity’s draft into the content queue as a pending item.
geo.status	geo_status	GET /geo	strategy:run	read	low	none	cli, mcp	—	Read current GEO (AI-citation) status: which AI answers mention the product.
geo.scan	geo_scan	POST /geo/scan	geo:run	spend	low	none	cli, mcp	product_id	Run a GEO scan to check whether AI assistants cite the product. Costs credits.
geo.checklist	geo_checklist	GET /geo/checklist	strategy:run	read	low	none	cli, mcp	—	Read the GEO readiness checklist (signal, status) plus which sources already cite the product.
geo.competitors	geo_competitors	GET /geo/competitors	strategy:run	read	low	none	cli, mcp	—	List competitor citation share and latest GEO visibility data.
geo.uncited	geo_uncited	GET /geo/uncited	strategy:run	read	low	none	cli, mcp	—	List high-value GEO queries where the product is not cited.
echo.mentions	echo_mentions	GET /echo/mentions	analytics:read	read	low	none	cli, mcp	—	List tracked brand and competitor mentions.
echo.scan	echo_scan	POST /echo/scan	analytics:read	spend	medium	approval	cli, mcp	—	Scan tracked sources for brand and competitor mentions.
voice.get	voice_get	GET /settings/voice	strategy:run	read	low	none	cli, mcp	—	Read canonical workspace brand voice.
voice.set	voice_set	PUT /settings/voice	settings:write	internal_write	medium	approval	cli, mcp	voice, guidelines, revision	Update canonical workspace brand voice with optimistic revision control.
audience.import	audience_import	POST /audience/import	workspace:write	internal_write	medium	approval	cli, mcp	csv*, consentSource*, source, enroll	Import subscribers from CSV with a required consent source.
recommendations.list	recommendations_list	GET /recommendations	strategy:run	read	low	none	cli, mcp	—	List the CMO’s current recommendations for the workspace.
recommendations.generate	recommendations_generate	POST /recommendations	strategy:run	internal_write	low	none	cli, mcp	—	Regenerate the recommendation set from the latest workspace data.
recommendations.decide	recommendations_decide	PATCH /recommendations/{id}	strategy:run	internal_write	low	none	cli, mcp	id*, action*	Accept or dismiss one recommendation. Accepting is what changes prioritization.
campaigns.list	campaigns_list	GET /campaigns	strategy:run	read	low	none	cli, mcp	—	List campaigns (multi-step marketing graphs) in the workspace.
campaigns.create	campaigns_create	POST /campaigns	campaigns:write	internal_write	low	none	cli, mcp	name*, graph	Create a campaign. Pass a node/edge graph to define its steps, or omit it for an empty draft.
campaigns.get	campaigns_get	GET /campaigns/{id}	strategy:run	read	low	none	cli, mcp	id*	Read one campaign including its full step graph.
campaigns.update	campaigns_update	PATCH /campaigns/{id}	campaigns:write	internal_write	low	none	cli, mcp	id*, name, status, graph	Update a campaign’s name, status (draft/active/archived), or step graph.
campaigns.archive	campaigns_archive	DELETE /campaigns/{id}	campaigns:write	delete	medium	approval	cli, mcp	id*	Delete a campaign.
workflows.start	workflows_start	POST /workflows/start	strategy:run	spend	low	none	cli, mcp	kind*, product_id*, limit, count	Start a background workflow (e.g. reddit_scan, hn_scan, content drafts) for a product. Costs credits.
workflows.execute	workflows_execute	POST /workflows/execute	strategy:run	spend	low	none	cli, mcp	workflowType*, input	Execute a named workflow synchronously with a small input map. Costs credits.
workflows.status	workflows_status	GET /workflows/{id}	strategy:run	read	low	none	cli, mcp	id*	Read a running workflow’s status and step results.
tasks.list	tasks_list	GET /tasks?status={status}&agent_id={agent_id}	strategy:run	read	low	none	cli, mcp	status, agent_id	List agent tasks, optionally filtered by status or agent.
tasks.create	tasks_create	POST /tasks	strategy:run	internal_write	low	none	cli, mcp	agent_id*, type*, input	Queue a task for one of the agents (scout, writer, publisher, engage, analyst).
tasks.get	tasks_get	GET /tasks/{id}	strategy:run	read	low	none	cli, mcp	id*	Read one task with its input, result, and error.
tasks.update	tasks_update	PATCH /tasks/{id}	strategy:run	internal_write	low	none	cli, mcp	id*, status, result, error	Update a task’s status, result, or error.
tasks.approve	tasks_approve	POST /tasks/{id}/approve	strategy:run	internal_write	medium	approval	cli, mcp	id*	Approve a task that is waiting for a human decision. Never approve your own work.
tasks.execute	tasks_execute	POST /tasks/{id}/execute	strategy:run	internal_write	medium	approval	cli, mcp	id*	Run an approved task now instead of waiting for the scheduler.
approvals.list	approvals_list	GET /approvals	queue:read	read	low	none	cli, mcp	—	List items waiting for a human approval decision.
approvals.approve	approvals_approve	POST /approvals/{id}/approve	publish	external_write	high	step_up	cli, mcp	id*	Record a human approval, which releases the item to publish. Agents must not call this on their own drafts.
approvals.reject	approvals_reject	POST /approvals/{id}/reject	queue:write	internal_write	medium	approval	cli, mcp	id*	Reject a pending approval so the item is never published.
launch.prepare	launch_prepare	POST /agents/launch	strategy:run	internal_write	low	none	cli, mcp	productId	Prepare launch assets (Product Hunt / Hacker News copy) for a product. Submission stays human-assisted.
listings.list	listings_list	GET /listings	strategy:run	read	low	none	cli, mcp	—	List directory launch listings (Product Hunt, Hacker News, directories) tracked for the workspace.
listings.create	listings_create	POST /listings	workspace:write	internal_write	low	none	cli, mcp	name*, url, status, notes	Track a new directory launch listing. Records intent only; submission stays human-assisted.
listings.update	listings_update	PATCH /listings/{id}	workspace:write	internal_write	low	none	cli, mcp	id*, name, url, status, notes	Update a tracked launch listing (status, URL, notes).
listings.archive	listings_archive	DELETE /listings/{id}	workspace:write	delete	medium	approval	cli, mcp	id*	Remove a tracked launch listing.
prospects.scan	prospects_scan	POST /prospect/scan	strategy:run	spend	low	none	cli, mcp	sources, keywords	Scan for prospects matching the workspace ICP. Costs credits.
inbox.list	inbox_list	GET /inbox	strategy:run	read	low	none	cli, mcp	—	List inbox conversations with prospects across email and DMs.
inbox.get	inbox_get	GET /inbox/{contactId}	strategy:run	read	low	none	cli, mcp	contactId*	Read one contact’s full inbox thread.
outreach.draft	outreach_draft	POST /inbox/send	outreach:write	internal_write	low	none	cli, mcp	contactId*, channel*, subject, body*	Draft an outreach message to a contact and queue it for human approval. Nothing is sent — approve and publish the queued item to actually deliver it.
crm.sync	crm_sync	POST /integrations/hubspot/sync	outreach:write	external_write	high	approval	cli, mcp	contactId	Push a contact to the connected HubSpot CRM. Writes to an external system, so it needs approval.
analytics.overview	analytics_overview	GET /analytics/overview?days=30	analytics:read	read	low	none	cli, mcp	days	Read aggregate performance over the last N days: totals per metric, per platform, and queue status.
analytics.attribution	analytics_attribution	GET /analytics/attribution	analytics:read	read	low	none	cli, mcp	—	Read attribution: which content and channels drove signups.
analytics.pagespeed	analytics_pagespeed	GET /analytics/pagespeed?url={url}	analytics:read	read	low	none	cli, mcp	url*	Run a PageSpeed check on a public url and return its performance scores.
analytics.report	analytics_report	GET /analytics/report?days=7	analytics:read	read	low	none	cli, mcp	days	Read the rolled-up performance report for the last N days (default 7).
feed.list	feed_list	GET /feed	strategy:run	read	low	none	cli, mcp	—	Read the agent activity feed — what the agents did and what needs attention.
feed.update	feed_update	PATCH /feed/{id}	strategy:run	internal_write	low	none	cli, mcp	id*	Mark one feed item as read or acted on.
ga4.pull	ga4_pull	GET /integrations/ga4	analytics:read	read	low	none	cli, mcp	—	Pull the latest metrics from the connected GA4 property.
github.repos	github_repos	GET /integrations/github/repos	connections:read	read	low	none	cli, mcp	—	List repos visible through the connected GitHub account, for shipping-log content.
connections.health	connections_health	GET /integrations/health	connections:read	read	low	none	cli, mcp	—	Check each connected platform’s token health before trying to publish.
agent_mode.get	agent_mode_get	GET /settings/agent-mode	strategy:run	read	low	none	ui, cli, mcp	—	Read the publishing mode (approve or autonomous) plus any per-agent overrides.
agent_mode.set	agent_mode_set	PUT /settings/agent-mode	settings:write	internal_write	high	step_up	ui, cli, mcp	global_mode, orchestrate_enabled, overrides	Change the publishing mode. Switching to autonomous removes the human approval gate, so it needs a human.
billing.status	billing_status	GET /billing/status	billing:read	read	low	none	cli, mcp	—	Read the active plan, credit balance, and renewal date.
me.get	me_get	GET /me	billing:read	read	low	none	cli, mcp	—	Read the authenticated identity: user id, token type, and granted scopes.
audience.list	audience_list	GET /audience?status={status}&segment={segment}	strategy:run	read	low	none	cli, mcp	status, segment	List the email marketing audience for the workspace, with status and segment filters.
audience.create	audience_create	POST /audience	workspace:write	internal_write	low	none	cli, mcp	email*, consentSource, source, segments	Add one subscriber to the email audience. A consent source is recorded for compliance (defaults to manual).
brief.get	brief_get	GET /brief	queue:read	read	low	none	cli, mcp	—	Read the most recent stored daily brief: yesterday’s shipped items, engagement, pending actions, and alerts.
creative.plan	creative_plan	POST /creative/runs	creative:run	spend	medium	approval	ui, cli, mcp	productUrl*, productId, feature, aspect	Derive a launch-video brief and storyboard. Creates an approval; does not render or publish.
creative.render	creative_render	POST /creative/runs/{runId}/render	creative:run	spend	medium	approval	ui, cli, mcp	runId*	Start rendering an approved storyboard. Heavy work runs on the media worker.
creative.status	creative_status	GET /creative/runs/{runId}	creative:read	read	low	none	ui, cli, mcp	runId*	Read a creative run, its stages, and signed preview URLs.
creative.review	creative_review	GET /creative/runs/{runId}/review	creative:read	read	low	none	ui, cli, mcp	runId*	Read the storyboard awaiting approval for a run.
creative.assets	creative_assets	GET /creative/runs/{runId}/assets	creative:read	read	low	none	ui, cli, mcp	runId*	List signed output assets for a completed creative run.
creative.cancel	creative_cancel	POST /creative/runs/{runId}/cancel	creative:run	internal_write	low	none	ui, cli, mcp	runId*	Cancel a creative run and refund unstarted stages.
creative.queue	creative_queue	POST /creative/runs/{runId}/queue	queue:write	internal_write	medium	approval	ui, cli, mcp	runId*, platforms	Create content_queue drafts from a completed creative run. Does not publish.
creative.models	creative_models	GET /creative/models	creative:read	read	low	none	ui, cli, mcp	—	List the active Creative Studio generation models with their credit cost per billing unit.
creative.generate	creative_generate	POST /creative/generate	creative:run	spend	medium	approval	ui, cli, mcp	prompt*, modelId*, kind, mode, duration, imageUrl, videoUrl, referenceUrls, targetPlatform, audio	Generate a Creative Studio image or video. Debits credits and drafts the result into the content queue for approval; never publishes.
creative.gallery	creative_gallery	GET /creative/gallery?kind={kind}&status={status}&cursor={cursor}&limit={limit}	creative:read	read	low	none	ui, cli, mcp	kind, status, cursor, limit	List Creative Studio generation jobs with signed output URLs, newest first.
creative.job	creative_job	GET /creative/jobs/{jobId}	creative:read	read	low	none	ui, cli, mcp	jobId*	Read one Creative Studio generation job and its signed output URL.
creative.abort	creative_abort	POST /creative/jobs/{jobId}/abort	creative:run	internal_write	low	none	ui, cli, mcp	jobId*	Abort an in-flight Creative Studio generation job; credits are refunded when the provider had not yet spent them.
creative.revise	creative_revise	POST /creative/runs/{runId}/revise	creative:run	spend	medium	approval	ui, cli, mcp	runId*, note	Open a new storyboard revision that supersedes a previous run. Does not render or publish.
adlib.search	adlib_search	POST /ad-library/search	discovery:run	spend	low	none	cli, mcp	query, platforms, country, limit	Collect competitor ads from the public ad libraries (Meta/Google/LinkedIn/TikTok/X) for a query, persist them, and return the scored page. Costs ad_library_scan credits; a live scan can take 15-40s.
adlib.browse	adlib_browse	GET /ad-library?q={q}&platform={platform}&competitor={competitor}&tier={tier}&board={board}&sort={sort}&limit={limit}&offset={offset}&min_runtime_days={min_runtime_days}&saved_only=1	discovery:read	read	low	none	cli, mcp	q, platform, competitor, tier, board, saved_only, min_runtime_days, sort, limit, offset	Browse ads already stored in the workspace’s ad library. Tiers are high_conf | winner | emerging | loser (not the winner_engine vocabulary).
adlib.get_ad	adlib_get_ad	GET /ad-library/{ad_id}	discovery:read	read	low	none	cli, mcp	ad_id*	Read one stored ad by its ad_id.
adlib.saved	adlib_saved	GET /ad-library/saved?board={board}	discovery:read	read	low	none	cli, mcp	board, limit	List saved ads with per-board counts.
adlib.save_ad	adlib_save_ad	POST /ad-library/saved	discovery:run	internal_write	low	none	cli, mcp	ad_id*, board, note	Save an ad to a swipe-file board (default board when omitted).
adlib.unsave_ad	adlib_unsave_ad	DELETE /ad-library/saved/{ad_id}?board={board}	discovery:run	internal_write	low	none	cli, mcp	ad_id*, board	Remove an ad from a board (default board when omitted).
adlib.alerts	adlib_alerts	GET /ad-library/alerts?unread_only=true	discovery:read	read	low	none	cli, mcp	unread_only, limit	Competitor ads seen for the first time — the launch-alert feed.
adlib.ack_alerts	adlib_ack_alerts	POST /ad-library/alerts/ack	discovery:run	internal_write	low	none	cli, mcp	ad_ids	Mark launch alerts read. Omit ad_ids to acknowledge everything unread.
adlib.competitors	adlib_competitors	GET /ad-library/competitors	discovery:read	read	low	none	cli, mcp	—	List the competitor watchlist merged with advertisers already seen in the library.
adlib.track_competitor	adlib_track_competitor	POST /ad-library/competitors	discovery:run	internal_write	low	none	cli, mcp	name*, platform, domain	Add a competitor name to the watchlist (idempotent).
adlib.untrack_competitor	adlib_untrack_competitor	DELETE /ad-library/competitors/{name}	discovery:run	internal_write	low	none	cli, mcp	name*	Remove a competitor from the watchlist.
adlib.sync_competitor	adlib_sync_competitor	POST /ad-library/competitors/{name}/sync	discovery:run	spend	low	none	cli, mcp	name*, platform, domain, country, limit	Track a competitor and refresh-pull its live ads now. Costs ad_library_scan credits; new ads raise alerts.
adlib.clone	adlib_clone	POST /ad-library/clone	discovery:run	internal_write	low	none	cli, mcp	ad_id*	Remix an ad’s hook into 3 deterministic variants for your own creatives.
ads.launch	ads_launch	POST /ads/launch	campaigns:write	external_write	high	approval	cli, mcp	platform*, ad_id, name, hook, body, cta, creative_url, persona, score, daily_budget, draft_copy, dry_run	Plan (dry_run=true, default) or execute an ad launch. Execution is policy-gated: winner score >= 60 creates a paused campaign draft, below that it lands pending_approval for human sign-off — nothing ever goes live directly.
ads.launches	ads_launches	GET /ads/launches?status={status}	analytics:read	read	low	none	cli, mcp	status, limit	List launched/paused/pending ad campaigns in the workspace.
ads.get	ads_get	GET /ads/launches/{id}	analytics:read	read	low	none	cli, mcp	id*	Read one ad launch by id.
ads.manage	ads_manage	PATCH /ads/launches/{id}	campaigns:write	external_write	high	approval	cli, mcp	id*, action, daily_budget	Manage one launch: action approve|activate|pause|archive (approve pending_approval -> paused; activate paused -> active; pause active -> paused; archive any -> archived) and/or set daily_budget.
ads.performance	ads_performance	GET /ads/launches/{id}/performance	analytics:read	read	low	none	cli, mcp	id*	Performance for one launch: spend, impressions, clicks, conversions, CTR, CPA, ROAS. synced:false + zeros until a platform metrics sync has written data.
autopilot.status	autopilot_status	GET /autopilot/status	autopilot:read	read	low	none	cli, mcp	—	Autopilot rollup: active loops, cycles run, pending proposals, learnings this week, plus every loop’s last-run state.
autopilot.learnings	autopilot_learnings	GET /autopilot/learnings?channel={channel}&kind={kind}&limit={limit}	autopilot:read	read	low	none	cli, mcp	channel, kind, limit	The readable learning files the loops append to each run — winner/miss/audit/digest lessons per channel with the rows that earned them.
autopilot.run	autopilot_run	POST /autopilot/run	autopilot:run	internal_write	low	none	cli, mcp	loop	Run one autopilot cycle now (or the weekly audit with loop=audit). Deterministic — grades signals, writes learnings, proposes actions. No credits; proposals still need approval.
autopilot.actions	autopilot_actions	GET /autopilot/actions?status={status}	autopilot:read	read	low	none	cli, mcp	status	List autopilot proposals — pending ones awaiting a decision, or applied/dismissed history.
autopilot.decide	autopilot_decide	POST /autopilot/actions	autopilot:write	external_write	high	approval	cli, mcp	id*, decision*	Decide one autopilot proposal: apply executes it (scale budget / pause launch / pause loop), dismiss closes it without side effects.
advocacy.list	advocates_list	GET /advocates?status={status}	advocacy:read	read	low	none	cli, mcp	status	Advocate leaderboard: repeat positive engagers scored on reach(momentum), loyalty, content and amplification, tiered micro_influencer/loyal_customer/ugc_creator/advocate.
advocacy.outreach	advocates_outreach	GET /advocates/{id}/outreach	advocacy:read	read	low	none	cli, mcp	id*	Ready-to-send DM template for one advocate, worded per tier (micro-influencer affiliate offer, loyal-customer early access, UGC feature request).
advocacy.set_status	advocates_set_status	PATCH /advocates	advocacy:write	internal_write	low	none	cli, mcp	id*, status*	Mark an advocate contacted / partner / dismissed once you’ve reached out.
sentiment.series	sentiment_series	GET /sentiment?days={days}	sentiment:read	read	low	none	cli, mcp	days	Daily per-platform sentiment rollup (positive/neutral/negative %) plus live anomalies vs the 7-day baseline — the crisis early-warning feed.

Scopes

Scope	Capabilities
advocacy:read	advocacy.list, advocacy.outreach
advocacy:write	advocacy.set_status
analytics:read	echo.mentions, echo.scan, analytics.overview, analytics.attribution, analytics.pagespeed, analytics.report, ga4.pull, ads.launches, ads.get, ads.performance
autopilot:read	autopilot.status, autopilot.learnings, autopilot.actions
autopilot:run	autopilot.run
autopilot:write	autopilot.decide
billing:read	billing.status, me.get
campaigns:write	campaigns.create, campaigns.update, campaigns.archive, ads.launch, ads.manage
connections:read	connections.list, reddit.status, github.repos, connections.health
creative:read	creative.status, creative.review, creative.assets, creative.models, creative.gallery, creative.job
creative:run	creative.plan, creative.render, creative.cancel, creative.generate, creative.abort, creative.revise
discovery:read	adlib.browse, adlib.get_ad, adlib.saved, adlib.alerts, adlib.competitors
discovery:run	adlib.search, adlib.save_ad, adlib.unsave_ad, adlib.ack_alerts, adlib.track_competitor, adlib.untrack_competitor, adlib.sync_competitor, adlib.clone
geo:run	geo.scan
outreach:write	outreach.draft, crm.sync
publish	queue.publish, approvals.approve
queue:read	queue.list, queue.get, queue.best_time, approvals.list, brief.get
queue:write	queue.delete, queue.update, queue.approve, queue.reject, queue.schedule, queue.proof, opportunities.queue, approvals.reject, creative.queue
sentiment:read	sentiment.series
settings:write	voice.set, agent_mode.set
strategy:run	strategy.run, scout.research, writer.generate, reddit.warmup, command.chat, command.sessions, command.run, command.status, command.approve_step, workspace.context, workspace.get, products.list, artifacts.list, listening.list, listening.scan, signals.scan, opportunities.list, opportunities.draft, geo.status, geo.checklist, geo.competitors, geo.uncited, voice.get, recommendations.list, recommendations.generate, recommendations.decide, campaigns.list, campaigns.get, workflows.start, workflows.execute, workflows.status, tasks.list, tasks.create, tasks.get, tasks.update, tasks.approve, tasks.execute, launch.prepare, listings.list, prospects.scan, inbox.list, inbox.get, feed.list, feed.update, agent_mode.get, audience.list
workspace:write	products.create, products.update, products.archive, audience.import, listings.create, listings.update, listings.archive, audience.create

User tokens (aicmo_usr_…, from aicmo login) carry all scopes. Agent tokens (aicmo_agt_…, from aicmo token create) carry an explicit subset. Capabilities with scope none are local and need no token.

Agent tokens currently reach only the native /api/v1/* handlers (queue, connections, workspace context, me), where withAuth enforces the declared scope. Capabilities served by a legacy /api/* route through the /api/v1 rewrite need a user token: those routes declare no scope, so a scoped agent token cannot be checked against one and is refused rather than silently granted everything. Declaring scopes on those handlers is what lifts the restriction.

Surfaces

chat capabilities are also AI CMO runtime tools under the same id. Every capability with an HTTP transport is reachable from the CLI as aicmo run <capability.id> '<json>' (run aicmo run with no id to list them); hand-written commands like aicmo queue approve are ergonomic aliases over the same registry entries.

Error envelope (all surfaces)

{ "error": { "code": "stable_snake_case", "message": "human readable", "status": 409, "retryable": false } }

MCP wraps it as { isError: true, content: [{ type: "text", text: "<that JSON>" }] }. Codes you will see: missing_auth_token, invalid_arguments, unknown_tool, untrusted_auth_override, untrusted_api_base_url, invalid_api_response, http_<status>, plus server codes such as reddit_not_connected, publish_not_approved, insufficient_credits, assisted_publish.

Live discovery endpoint: GET https://app.aicmohq.com/api/v1/capabilities.