Feature / MCP
Give agents curated Brandloop tools, not provider credentials.
Brandloop's MCP foundation is designed for controlled marketing automation: submit brand sources, upload or read assets, create or inspect generation jobs, and keep agents inside product-level contracts.
What MCP can safely automate
Submit brand sources
Agents can hand Brandloop a public website or social profile for brand intelligence intake through a scoped contract.
Work with assets
Asset upload and retrieval tools support brand references, generation inputs, outputs, and knowledge attachments.
Create generation jobs
Generation tools can request Brandloop media or content jobs while staying behind API-key scopes and idempotency boundaries.
Why MCP belongs behind Brandloop
- Small tool surface
- The tool list is intentionally narrow so agents cannot mutate billing, admin settings, provider credentials, or destructive resources.
- Same API contracts
- MCP uses Brandloop product contracts instead of creating a separate provider-specific path. The SDK it consumes is an internal monorepo preview, not a public package.
- Workspace scoping
- API-key authorization keeps calls tied to the account and workspace behind the key.
- HTTP endpoint foundation
- The app includes an `/api/mcp` route with bearer authentication and MCP protected-resource metadata handling.
MCP vs generic agent automation
| Alternative | Where it breaks | Brandloop approach |
|---|---|---|
| Raw browser automation | Powerful, but easy to drift outside approved product and workspace boundaries. | Brandloop MCP exposes curated operations that map to the product contracts. |
| Provider API keys in agents | Agents can inherit credentials they should never see. | Brandloop keeps provider secrets behind first-party API and MCP surfaces. |
| Unscoped internal scripts | Hard to audit and often bypass review or idempotency. | Brandloop MCP and API operations are designed around scoped keys and explicit tool metadata. |
Current claim boundary
- MCP is a foundation for controlled automation. Autopilot plans campaigns on a cadence, but it publishes nothing unattended — every draft waits on human approval.
- The MCP server intentionally avoids billing, admin, provider-secret, and destructive tools.
- Remote MCP supports OAuth 2.1 and API keys, but it is not yet a one-click directory-listed connector in Claude or ChatGPT.
Common questions
- Which MCP tools are exposed?
- The current curated surface exposes whoami, brand_source_submit, brand_source_get, asset_list, asset_upload, asset_upload_complete, asset_get, generation_profiles_list, generation_create, and generation_get.
- Can MCP change billing or provider settings?
- No. The MCP package explicitly avoids billing, admin, provider-secret, and destructive tools.
- How does MCP authenticate?
- The remote endpoint accepts either an OAuth 2.1 access token, discovered and obtained by the client itself through PKCE and dynamic client registration, or a Brandloop API key as a bearer token. The local stdio server requires BRANDLOOP_API_KEY and BRANDLOOP_BASE_URL.