Organization-bound access
Choose one organization during consent. Every tool inherits that boundary without exposing a tenant selector.
Give Codex a controlled path into your social workspace
BlendDuck exposes an OAuth-protected MCP server so an AI client can work inside one selected organization, read publishing state, and create posts within explicit scopes.
- 20
- curated core tools
- 20
- opt-in API tools
- 5
- resource categories
Twenty-one task-oriented tools, kept intentionally curated
The default server covers posts, media, channel health, scheduling, analytics, inbox, and approvals. Agents may separately authorize 20 OpenAPI-derived resource tools in five categories without loading the entire API surface.
list_channels
Inspect connected channels and whether each destination can currently publish.
list_posts + get_post
Review posts across an explicit date range and inspect one post's current publishing state.
get_analytics + get_inbox
Read normalized performance evidence and the actionable engagement queue without sending or moderating.
create_post
Create a draft, schedule it, or request publishing to one authorized channel.
api_* resource tools
Load only authorized workspace, channel, post, media, provider, or webhook operations derived from the reviewed v1 contract.
Natural language does not bypass product security
MCP uses the same Better Auth identity and workspace authorization model as the dashboard.
OAuth discovery and PKCE
Clients discover the authorization server, register dynamically, and complete a PKCE-protected login.
Readable consent
Users see whether the client is requesting workspace reads or post creation before access is granted.
Audience and scope checks
JWT audience, issuer, user subject, and blendduck:read or blendduck:write scopes are validated server-side.
The agent operates the product, not a shadow database
MCP calls the same audited application use cases and social publishing flow used by native product surfaces.
Shared workspace state
Connected accounts, posts, roles, and publishing results stay consistent across agent and dashboard workflows.
Structured tool results
Tools return machine-readable content and explicit errors that clients can inspect before taking another action.
Expandable without stubs
New tools are added only when the underlying product capability and authorization path actually exist.
Let Codex see the same publishing state your team sees
Add the hosted MCP endpoint, complete OAuth, and start with read access before granting post creation.