Browse all guides
All guides
MCP

MCP tool reference

Inputs and examples for every Tartol MCP tool.

12 min readUpdated August 27, 2026

Quick visual guide

Connect first, then check live capabilities before calling a tool.

  1. 1

    Connect the client

    Use the client-specific instructions in Settings > Integrations.

    MCP client setup tabsPick your client
  2. 2

    Check capabilities

    Call whoami and list_capabilities before assuming a tool is available.

    MCP guide and reference navigationTool reference
  3. 3

    Use the documented action flow

    For mutations, describe or prepare the action first, then execute only after approval.

    MCP documentation detailsFollow the contract

Red labels show the exact control to use.

Detailed referenceOpen this for definitions, limits, examples, and troubleshooting.

Tool list

ToolRequired scopePurpose
whoamiAny valid connectionReturn the workspace, role, authentication method, and scopes.
list_capabilitiesAny valid connectionList the views and actions available to this connection.
searchtartol:readFind records and return stable tartol:// IDs.
fetchtartol:readRead one resource by a stable ID.
query_workspacetartol:readRead an available workspace view with filters.
get_generation_resulttartol:readPoll image, video, or copy generation and return completed media previews.
describe_actionWrite or GenerateReturn the schema and example for one available action.
execute_actionWriteValidate and run an action that does not spend workspace credits.
prepare_actionWrite or GenerateValidate a credit-spending or independently approved external action and return its exact confirmation summary.
execute_prepared_actionWrite or GenerateRun one explicitly approved prepared action.

Supported Tartol workflows

Tartol MCP is Tartol’s conversational action surface for compatible external AI clients. Tartol Agent is the separate in-app guidance surface: it explains workflows and troubleshooting but does not perform actions.

Tartol MCP focuses on bounded workflows that are useful and understandable in an AI conversation. It does not mirror every internal or visual control in the web app. The underlying Tartol feature can remain available in the app even when it is intentionally excluded from MCP.

  • Research and fetch Discovery ads, saved ads, brands, products, creatives, copy, tracked brands, and media.
  • Organize Saved Ads, Brand Tracker, product strategy, and Media Library records without exposing structural or cascading deletion.
  • Create and update brands, products, audiences, and marketing angles.
  • Use describe_action for the current standardized productType and billingPeriod values; free-form product categories are rejected by catalog write actions.
  • Generate, edit, translate, organize, and save image and copy results with the required credit approval flow.
  • Read the Ad Manager overview/history and fixed 30-day performance view, then prepare and execute an exact saved draft as Paused ads after independent approval.

whoami and list_capabilities

whoami arguments
{}
list_capabilities arguments
{
  "action_offset": 0,
  "action_limit": 20
}
ToolInput fields
whoamiNo input fields.
list_capabilitiesaction_offset: integer ≥ 0. action_limit: 1–25, default 20.

fetch

Returns one current-workspace resource. The ID must come from search, fetch, or query_workspace.

Arguments
{
  "id": "tartol://product/123"
}

query_workspace

Reads one view listed by list_capabilities. view is required and limit is 1–12. Available views depend on the connection.

Find long-running EU ads
{
  "view": "discovery_ads",
  "preset": "eu_winners",
  "sort": "longest_running",
  "running_days_min": 14,
  "limit": 12
}
Read Ad Manager overview
{
  "view": "meta_ads_overview"
}
Read 30-day Meta performance
{
  "view": "meta_ads_performance"
}
Filter groupFields
Generalview, search, limit
Ad ordersort: newest, oldest, or longest_running
Catalogcatalog_kind, product_id
Ad attributesplatform, format, status, niche, product_category, cta, text_presence
Discovery presetsready_to_scale, native_ads, eu_winners
Rangesdescription_length_min/max, running_days_min/max
EU metricseu_only, eu_views_min/max, eu_spend_min/max
Page metricspage_reach_min/max, page_spend_min/max, page_active_ads_min/max, page_total_ads_min/max
Identifiersplatform_ad_id, folder_id

Media previews and asynchronous results

Tartol returns important images, logos, product art, creatives, ads, and screenshots as native MCP image content with an exact detected MIME type, in addition to structured data. This lets compatible AI clients render the image bytes directly instead of trying to preview an external resource URL. Video and other unsupported or oversized media remain typed resource links because MCP does not define a native video content block and tool responses must stay bounded.

  • After scan_brand_from_url or rescan_brand, poll fetch with the returned brand resource_id until scanStatus is completed or failed. The completed brand response includes its products and their media.
  • After generate_image or generate_copy starts, poll get_generation_result with the returned job ID until completed or failed.
  • Present the important record details and every attached native image or media resource link. A client that cannot embed a media type should still expose its clickable link.
Poll image generation
{
  "kind": "image",
  "job_id": 456
}
Poll copy generation
{
  "kind": "copy",
  "job_id": 789
}

Action execution and approval

Call describe_action before acting. It returns the current JSON Schema, constraints, example, approval_required value, and predictable cost inputs where available. Most actions that do not spend workspace credits use execute_action directly. Credit-spending actions and independently approved external actions use the prepare, review, approve, and execute flow. If a free prerequisite such as a new-brand scan comes before image generation, quote the downstream image cost and ask for approval before starting the scan.

For generate_product_audiences and generate_product_angles, count is required from 1 to 8 even though Tartol’s UI defaults to 1. Each item costs 5 credits, and an audience includes its background portrait. The Product Strategy view returns generation_guidance plus audiences.data[].id. To use one for Image Ads, set generationConfig.autoAudience to false and generationConfig.selectedAudienceId to that numeric ID. Never send avatarUrl or avatarStoragePath; Tartol resolves eligible portrait evidence after validating ownership.

For launch_meta_ads_draft, copy draftId and draftUpdatedAt from meta_ads_overview. prepare_action revalidates the exact saved draft and returns a paused-only summary. After a person approves it, execute_prepared_action rechecks freshness and publishes. Use meta_ads_performance for the fixed 30-day selected-account report. MCP cannot connect Meta, change defaults, create campaign structures, change delivery, choose custom report controls, import provider links, submit unsaved rows, or activate ads.

Describe
{
  "action_type": "create_saved_ad_board"
}
Execute a non-credit action
{
  "action_type": "create_saved_ad_board",
  "arguments": {
    "name": "Strong product demos"
  }
}
Prepare a 10-credit brand product scan
{
  "action_type": "detect_brand_products",
  "arguments": {
    "targetId": 123
  }
}
Prepare a paused Ad Manager draft
{
  "action_type": "launch_meta_ads_draft",
  "arguments": {
    "draftId": "00000000-0000-4000-8000-000000000000",
    "draftUpdatedAt": "2026-08-14T12:00:00.000Z"
  }
}
Execute after approval
{
  "pending_action_id": "PASTE_RETURNED_ID",
  "confirm": true
}
ToolRequired fields
describe_actionaction_type
execute_actionaction_type, arguments
prepare_actionaction_type, arguments
execute_prepared_actionpending_action_id, confirm=true

Resources and prompts

TypeName or URIPurpose
Resourcetartol://workspace/currentCurrent connection and workspace context.
Resourcetartol://capabilitiesAvailable views and actions.
Resourcetartol://guideMCP usage and safety guidance.
Promptworkspace_briefSummarize workspace activity and performance.
Promptad_researchResearch ads using current Tartol data.
Promptbrand_monitoring_reviewReview monitored brands and follow-ups.

Availability

Tools, views, and actions can vary by scope, workspace role, plan, and feature access. Intentional MCP exclusions remain unavailable regardless of role. Call list_capabilities at the start of a session instead of keeping a fixed action list.

Keep learning

Still need help?

Tell us what you were trying to do and what happened.

Contact support
Last updated August 27, 2026