Documentation
Back to Dashboard

BlackOps Center OpenAI GPT Integration

Use BlackOps from a custom GPT with Actions so your GPT can read context, draft content, and run publishing workflows with your account token. The action schemas are generated directly from the BlackOps tool registry, so they stay in lock-step with what the MCP server exposes to Claude.

Two ways to connect

ChatGPT caps each Action at roughly 30 operations, while the full toolset is larger than that. Pick based on how much you need:

  • Quick start — import one curated Action (28 of the most common publishing operations: notes, posts, tweets, reservoirs). Fits in a single Action with room to spare.
  • Full coverage — import several domain-grouped Action sets (posts, social, media, brains, video, and more) as separate Actions in the same GPT for the complete surface.

Quick start — Publisher GPT

  1. In ChatGPT, create or edit a custom GPT and open the Actions section.
  2. Add an Action and import this schema URL:
https://www.blackopscenter.com/.well-known/blackops-gpt-openapi-subset.yaml
  1. Set Authentication to API Key → Bearer, and paste a BlackOps API token (generate one under Settings → API tokens).
  2. Save and test with a simple call, for example: list posts by domain.

Full coverage — multiple Action sets

Each domain group is published as its own schema under /.well-known/. Add each one you need as a separate Action in the same GPT, with the same bearer token. The authoritative list of available schema files and their operation counts is in the manifest:

https://www.blackopscenter.com/.well-known/blackops-gpt-openapi-manifest.json

Each entry's file resolves to https://www.blackopscenter.com/.well-known/<file>. A complete single-file bundle (useful for tooling or review, not for import — it exceeds the per-Action cap) is at /gpt-actions/openapi.full.yaml.

Authentication

  • Header format: Authorization: Bearer <token>
  • Tokens are scoped to your sites. The GPT chooses which site to act on from domain or site_id parameters on each call.

Prompting pattern

For reliable behavior, direct the GPT to resolve site context first, then perform content operations with explicit site parameters:

  1. Call get_me to discover accessible sites. If only one site exists, use it; if several, ask the user which one.
  2. Optionally call get_brand_voice / get_ai_context before drafting.
  3. Pass domain or site_id on every read/write so calls are tenant-scoped.

Publishing safety

Destructive and publish actions require "confirm": true in the request body (publishing a post, posting a tweet, deleting records). Keep an explicit confirmation step in your GPT instructions so nothing goes live without operator approval.

What's not available via GPT Actions

Google Analytics (GA4) and Search Console (GSC) tools are MCP-only. They run against Google's APIs from inside the MCP server and have no REST endpoint a GPT Action can call, so they are intentionally excluded from these schemas. To use analytics, connect via the MCP server (the same one Claude uses) from an MCP-compatible client — see Claude CoWork MCP.

Related docs