What it can do

The tools below are what the assistant has available. You don't call them by name — you ask for what you want and the assistant picks. They're listed so you know what is in scope.

Integrate

Local mode only — these read and write your repository.

Tool	What it does
detect_stack	Scans your project and recommends an integration approach
add_sdk_snippet	Inserts the SDK for script-tag and npm projects
advise_gtm	Exact Google Tag Manager configuration
advise_segment	Exact Segment destination configuration
advise_posthog	Exact PostHog webhook or S3 export configuration
get_integration_guide	Links the right SDK integration page

detect_stack and get_integration_guide work without a connected account.

Verify

Tool	What it does
check_events_arriving	Confirms data reached Moveo One in the last N minutes
watch_events	Waits for the first event after you deploy
get_watch_status	Checks on a watch

This is the loop worth having an assistant for: change the code, deploy, and have it confirm events arrived — rather than switching to a dashboard and refreshing.

Analyze

Tool	What it does
get_overview	Session, event and participant counts for a period
list_dashboard_plots	Your dashboard plots
get_plot / refresh_plot	Read a plot, or recompute it for a date range
ask_analytics	A natural-language question over your data

ask_analytics is the same analytics agent the web app uses, so it answers with the same data and the same caveats.

Simulate

See Quantum for what these actually do.

Tool	What it does
list_cohorts / get_cohort	Your cohorts
create_cohort / delete_cohort	Create and remove a cohort
discover_for_cohort	Discover the target's structure for a cohort
calibrate_cohort / get_calibration_status	Start a calibration and follow it
run_simulation	Launch a software run — including A/B arms and debug capture
get_simulation_status	Progress, and the results summary when it finishes
list_runs / cancel_run	A cohort's runs; cancel one
list_sessions	The agent sessions in a run
get_session_screenshots	What one agent saw
get_session_debug	Why one agent acted — page state, model response, layer scores, winning action. Needs the run to have been launched with debug capture
create_apk_upload_url	An upload URL for an Android build
upload_app_binary	Uploads an APK from your machine — local mode only

Calibration

Tool	What it does
upload_calibration_data	Upload a dataset and start the fit
get_calibration_mapping / confirm_calibration_mapping	Review and confirm the column mapping — the required human gate
cancel_calibration	Stop a running calibration
list_calibrations	Your calibration layers, or one layer's detail and fit report
set_calibration_chain	Set which layers a cohort composes

Content track

Tool	What it does
get_content_status	Whether content testing is enabled, and the available media environments
enable_content_testing	Turn it on for the account
create_content_cohort	A cohort for creative testing
upload_creative	Upload a video, image or display asset
run_content_simulation	Launch a content run
get_creative_results	The creative's results

Archetypes and audiences

Tool	What it does
list_archetypes	The archetype library for either track
manage_software_archetype	Create, update or delete a software archetype
suggest_audience / suggest_archetype	Propose a mix from a plain-English brief
list_saved_audiences / manage_saved_audience	Reusable audiences
create_cohort_from_document	Build a cohort from an ICP or research document — local mode only

Simulations take hours to days

run_simulation returns immediately with a run id. The assistant polls get_simulation_status; it is not blocked waiting, and neither are you.

Models

Tool	What it does
list_prediction_models	Your models
get_prediction_model	One model's configuration
get_prediction_model_metrics	Its metrics
check_data_readiness	Whether you have enough data to train at all — the honest first step
list_training_config_options	Target events and grouping keys you can build a config from
validate_training_config	Check a config before spending a training run on it
request_model_training	Start training
list_training_requests / get_training_request	Follow it to completion
get_model_ab_stats	A/B statistics for a live model

Training takes hours, and a Readiness Scan allows one model

request_model_training starts a real job. Run check_data_readiness first, and get_plan_status if you're not sure how much of your quota is left. A finished training request is not yet a servable model — get_training_request reports that separately.

See Build a model for what the configuration means.

Experiments

Tool	What it does
list_experiments	Experiments, filterable by model
create_experiment	Create one — note it goes live on creation
get_experiment_results	A/B stats, periods and actions in one answer
update_experiment	Split, start date, enabled
test_experiment_webhook	Dry-run through the loopback, or deliver for real

See A/B testing.

Warehouse integrations

Tool	What it does
get_warehouse_status	What is wired up in both directions, and whether it is healthy
test_warehouse_connection	Test before you save
configure_warehouse	Create or update a configuration
set_warehouse_enabled	Turn one on or off

Data out is warehouse export; data in is Redshift or BigQuery import. Every tool states which direction it is acting on. Credential values are never returned.

Account

Tool	What it does
whoami	The connected account and reachable projects
list_projects	Workspaces you can access
get_plan_status	Your plan, days remaining, whether it has expired, and quota used
list_api_tokens / create_api_token	Personal tokens for headless clients — see Install
connect / connect_finish / logout	The browser device flow — local mode only

Guided prompts

Two prompts walk through the common flows end to end. Your client surfaces them however it surfaces MCP prompts — usually a slash command or a menu.

• onboard — from an uninstrumented project to events arriving.
• test_my_app — from a working integration to a Quantum simulation result.

Resource

• moveo://getting-started — orientation the assistant can read without asking you.

Scope

The MCP is a thin client over the same API the web app uses. It can do what your account can do. It does not talk to the database, it does not change your plan, and it does not delete models, experiments or integration configurations — those stay in the browser where a confirmation dialog exists.