Kaminari Ad

MCP server

Connect Claude and other AI agents to Kaminari Ad through the hosted Model Context Protocol server.

Kaminari Ad ships a Model Context Protocol (MCP) server so AI agents — Claude and any MCP-compatible client — can operate your account with natural language while every action goes through the same authenticated, permission-checked API.

What it exposes

The server exposes the Kaminari Ad operations as MCP tools: create and read scans, manage campaigns and campaign groups, list and triage alerts, manage tags, custom rules, taxonomies, policy sets, webhooks, billing reads, and more. Each tool maps to a REST endpoint and respects the caller's permissions.

Alerts can be triaged in bulk as well as one at a time, and the same filters the alerts list accepts — policy set, tag, country, date range — narrow both the listing and the counts, so an agent can size a selection before acting on it. Policy-set membership is editable incrementally, without replacing the whole campaign list. Scan artifacts are fetched through the server too: the screenshots as images, the creative markup and VAST document as text, and the MP4 as an attachment.

create_campaign, update_campaign, create_scan and create_bulk_scans take repeat_count, repeat_mode and retry_max_attempts with the same meaning and ceilings as the REST fields, shared + ad discovery included: the API answers 422 and the agent sees the error. On the way back, the two create tools return repeat_scan_ids — the sibling ids, which nothing else exposes — get_campaign echoes the three settings, and get_scan / list_scans add repeat_index, repeat_total, repeat_session_id, retry_attempt and retry_max_attempts. So an agent that asked for ten repeats can find the other nine and tell which attempt each one finished on.

An agent drafting detection rules can preview them before saving: test_custom_rule runs one rule against one scan, and test_custom_rules_batch runs up to 20 rules against up to 50 stored scans in one call, with the same cell errors and time limit as the Rule testing API.

Hosting and install

  • Hosted — https://mcp.kaminari.ad over HTTP, authenticated per request with a Bearer token (an API key or an OAuth access token).
  • Local / npm — the server is published as @kaminari-ad/mcp for stdio use; point it at your API key via environment variables.

Auth

Hosted MCP uses the same OAuth surface as connected apps (including RFC 9728 protected-resource metadata) so directory integrations can discover and authorize it. For local use, a single API key in the environment is enough.

When to use it

Reach for MCP when you want an agent to drive Kaminari Ad interactively. For deterministic, high-volume automation, call the REST API directly.

← All documentation