The Social Bee app for ChatGPT: what it can and cannot do
· Product
It reads your workspace, drafts and makes creative. It never publishes without an approval step, and it publishes to the exact connected account you chose.
The app reads your Social Bee workspace, plans, makes creative, writes drafts, moves work through approval where your role allows it, schedules, and publishes to the exact connected account after you confirm the action in the conversation. It never publishes without an approval step. It cannot grant access, invent a brand, bypass a permission or reveal a credential.
Social Bee's app for ChatGPT is an OAuth-protected Model Context Protocol server. A registered Social Bee user signs in with their own account, and every request then resolves the same tenant, brand, role, plan, allowance and connected-account context the product uses. It is an additional interface to Social Bee, not a bypass around its approval controls. If the server cannot establish the scope a request needs, it fails closed rather than guessing.
The tool surface is grouped around real workflows rather than a generic assistant abstraction: workspace context, planning, campaigns, product-led creative, creative production, the content workflow, connections and reporting, and the Instagram community inbox. The reviewed submission packet contains 52 tools. The runtime count is deliberately not published as a fixed promise, because it changes with releases: a release check on 8 September 2026 counted 61 tools, which is a dated reading rather than a contract. The OpenAI app itself was approved as version 1.0.1 on 24 August 2026.
What it can do: read the permitted brands, Brand Kit, allowances, connected platforms and operating context for the signed-in user; read and create events, trends, competitor context, guidance and post briefs; request image generation, read job status, browse approved and pending assets and attach a visual to a draft; create and edit drafts and request or record approvals where the role and plan allow; prepare an exact-account publish and, after your confirmation, execute it; read retained Instagram conversations and comments and prepare replies or moderation drafts; and read analytics and reports for permitted accounts.
What it cannot do, however the conversation is worded: grant access to a person who was not invited, or create a brand that did not exist; override the platform policy that governs the workspace, or widen a role; publish a generated visual automatically, or treat a pending asset as approved; skip an approval that applies in the product; publish to a platform name, or to an account you did not select; send a reply or perform moderation without your explicit confirmation of brand, account, target and action; or present staged or demo data as live, or reveal provider credentials.
The four stages are what make that second list hold. Read tools retrieve permitted state. Prepare tools create reviewable internal state and contact no provider. Approve tools advance an exact version through the Social Bee workflow when the role and plan allow it, and approval alone does not send a provider action. Execute tools can contact an external provider only after the user explicitly confirms the named brand, account, target and exact action in the current conversation, and the server independently validates the scoped, short-lived confirmation challenge and an idempotency key. That separation is what makes safe replies, comment moderation and publishing possible, and it is what makes a reliable audit trail possible afterwards.
Exact-account context is the other half. Provider reads and writes are scoped to an exact selected social account, never to a vague platform label. A user with access to more than one brand must be told which brand is active, and sensitive calls require an explicit brand selection. Before a community or publishing action, the active brand and the exact account are named. If the account pair changes or a scope is missing, the action fails closed and asks for a reconnect rather than finding a different account to use.
Some tools return Social Bee widgets so the conversation can show workspace, post or creative status without exposing raw internal APIs. Widgets are presentation, not authority: an asset shown as "Awaiting approval" may be viewed but cannot be treated as approved, scheduled or published.
One more guarantee, because it is where AI tools usually cut a corner. When you give the app a public product URL, an exact creative request has to retain a verified product identity and source image. If that proof is absent, the job fails honestly instead of substituting a lookalike or quietly turning the request into text-only generation. A successful import or generated image stays a draft or pending asset until the normal review workflow advances it, not posted merely because a conversation created it.
Costs behave the same way in chat as in the product. Image generation reserves an Image Generation Credit before the job and settles it against the delivered result; video reserves a matching prepaid clip unit. Asking through ChatGPT does not make generation free.
What the app does not decide is channel availability. Instagram feed posts, carousels, Reels, comments, and DMs are available for connected accounts. Facebook Pages connect today; Page publishing opens when Meta completes its review. Personal LinkedIn profiles publish today; Company Page publishing runs through the official API and reaches customer Pages as LinkedIn extends our access. TikTok posts privately until TikTok finishes reviewing the app. Snapchat is coming soon pending review. X is coming as a paid add-on channel; it is not available yet.
Connecting is deliberately boring. The canonical MCP resource is socialbee.hyve-os.com/mcp, the server uses dynamic client registration, authorization codes with PKCE and short-lived registered-user tokens, and a browser page or demo never grants access by itself. Do not paste a password, a Social Bee token, a Meta token or an API secret into a connector URL or a conversation. The same server can be added in Claude Desktop as a custom connector, against the same URL and the same OAuth boundary. Connecting a client does not connect an Instagram or LinkedIn account: that is done once, in the product, by someone who administers the account.
MCP access compatible with ChatGPT and Claude is included on every plan: Creator at $100 a month, Starter at $400, Growth at $1,000, and Enterprise from $1,499 with sales-assisted setup. Creator and Starter start with a 7-day trial: a card at checkout, the first charge on day 8, one trial per business, capped at 2 Image Generation Credits, 20 publishing actions and 2 connected profiles.
The short version: the conversation is a faster way into the same governed workspace. Every gate that exists in Social Bee still exists in the chat, and the one that matters most is the approval step before anything reaches an account you named.