Connect AI Assistants

What is MCP?

The Model Context Protocol (MCP) is an open standard that lets AI assistants call external tools. PostRite exposes a set of tools — creating posts, scheduling, listing connected accounts, and more — that a compatible assistant can use on your behalf. The connection is authorized with OAuth 2.1, the same family of standards banks and identity providers use, so the assistant never sees your password and only gets the access you approve.

Enabling AI assistants for your organization

For security, AI-assistant access is off by default for every organization. Before an assistant can connect, the organization owner must enable it under Settings → Organization → AI assistants. There are two independent switches:

• Remote MCP access — allows assistants like Claude to connect through the MCP server (the flow described below).
• In-page browser agents (WebMCP) — exposes in-page tools to browser AI agents while a member is using the PostRite app.

Turning a switch off takes effect immediately: pending consents are rejected, existing connections stop working on their next request, and their tokens can no longer be refreshed.

Remote MCP access has an extra safeguard: Allow deleting via remote MCP is a separate switch and is off by default. While it is off, an assistant connected through MCP can read and create, but any attempt to delete a post, template, or media file is refused — even if it holds the matching write scope. Turn it on only when you want assistants to be able to delete on your behalf; toggling it also takes effect immediately.

Enlarge

Connecting Claude

The quickest way is the Connect to Claude button under Settings → Organization → AI assistants: it opens Claude with PostRite already filled in. Click Add, and Claude opens the PostRite sign-in and consent screen in your browser:

1. Sign in to your PostRite account (if you are not already signed in).
2. Choose the organizations the assistant can act in — one or several. It can never reach the organizations you leave unchecked; to add one later, connect again. Only organizations whose owner enabled MCP access are selectable.
3. Review the requested permissions (scopes) and click Authorize. You are returned to Claude, now connected. The connector works in Claude on the web, desktop and mobile. The screen also says who vouches for the assistant — for Claude, identified by claude.ai. An app PostRite cannot verify is marked as not verified: authorize it only if you started the connection yourself.

You can also add it by hand: in Claude, add a custom connector and paste the MCP server URL above. On Claude Team or Enterprise, an organization owner adds it once for everyone — the link under the button opens that dialog.

Connecting ChatGPT

ChatGPT connects to PostRite as a custom MCP app through its Developer mode, which OpenAI offers on eligible paid plans. OpenAI moves these menus from time to time, so its help article on Developer mode has the current path:

1. In ChatGPT's settings, turn on Developer mode. On a Business or Enterprise workspace, an admin may have to allow it first.
2. Add a new app with the + button, name it (for example PostRite) and paste the MCP server URL above as its connection.
3. Choose OAuth as the authentication. ChatGPT then opens the same PostRite sign-in and consent screen described for Claude: pick the organizations and approve the permissions.
4. In a new chat, turn PostRite on from the tools menu.

Other MCP clients

Any client that supports remote MCP servers (Streamable HTTP) with OAuth 2.1 can connect the same way: add a remote server with the URL above and complete the PostRite consent screen. A client that only runs local servers cannot reach it.

What the assistant can do

Every request the assistant makes runs with the exact same permissions your own account has in that organization — it can never do more than you can. Available tools include:

• Organizations — list the organizations this connection can act in.
• Posts — list, view, create, schedule, update, cancel, reschedule, and retry posts. The assistant can validate a post against every platform's rules before creating or scheduling it (a dry run that saves nothing), and drafts it saves come back with advisory warnings listing anything to fix before scheduling. It can write different text per platform, set each network's own options (see below), send a post for external approval, resend an approval request and read a post's approval history.
• Templates — list, view, create, update, and delete reusable content templates, including a different body or different media per social network.
• Connected accounts — list your connected social accounts and whether any needs reconnecting, check platform availability, and look up what a post needs from a specific account: your Pinterest boards, the Discord channels the bot can post in, and the privacy levels your TikTok account allows.
• Media — list media, import an image or video from a URL, attach library media to posts by its ID, set or clear a file's alternative text, and delete media.
• Analytics — read a post's engagement metrics, the organization's analytics summary, top posts by engagement, and posting insights (best days and times, content types, cadence) from your last 12 months — for the whole organization, one platform, or one connected account. It can also turn your best time slots into concrete times to schedule at, skipping slots that already hold a post.
• Billing — read your plan, usage, and limits.

When a connection covers more than one organization, every tool takes the organization to act in, and the assistant asks you which one, or uses the one you named — it never guesses. A request aimed at an organization you did not select is refused.

Options per network

When it creates or updates a post, the assistant can set the same options the editor offers:

• Facebook — a feed post, a Reel or a Story; a first comment.
• Instagram — a post, Reel or Story; first comment, collaborators, trial Reels, cover, music, the AI label and paid partnership.
• X — who can reply (everyone, accounts you follow, or only accounts you mention); a first reply.
• Threads — who can reply and a first reply, plus sharing to your Instagram story and reply approvals when your organization has them enabled.
• YouTube — title, privacy, category, tags, made for kids, Shorts and thumbnail.
• TikTok — privacy level (chosen from the ones your account allows), comments, duets and stitches, branded content, the AI label, cover and photo title.
• Pinterest — board (the assistant can list yours), title, destination link and cover.
• LinkedIn — a custom thumbnail for videos.
• Google Business Profile — an update, event or offer, with a call-to-action button.
• Discord — the channel on each server, which the assistant can look up, or the webhook's name and avatar.
• Bluesky — who can reply and quote, and content labels.
• Mastodon — visibility, content warning and language.
• Telegram — silent delivery and link previews. WhatsApp — the recipient.

Things to ask

Plain requests are enough — the assistant looks up accounts, IDs and times itself. A few to start with:

Understanding scopes

Scopes are fine-grained permissions. During consent you can see exactly which are requested; an assistant only receives the scopes you approve, and each one is enforced on every request. What each one unlocks:

• posts:read — view posts and their approval history, plus everything in analytics: engagement, summaries, top posts, posting insights and suggested times.
• posts:write — validate, create, schedule, update, reschedule, cancel, retry and archive posts, and resend approval requests.
• templates:read / templates:write — view / manage templates.
• media:read — view the media library. media:write — import media from a URL, set alternative text and delete media.
• platforms:read — view connected accounts, their health and platform availability, and look up Pinterest boards, Discord channels and TikTok posting options.
• billing:read — view plan, usage and storage (organization owners only).

The Allow deleting via remote MCP organization setting applies on top of these scopes: even a token that holds posts:write, templates:write, or media:write cannot delete anything while that setting is off.

Revoking access

You can revoke an assistant at any time in PostRite, under Settings → Account → Connected AI assistants: every assistant you authorized is listed there, with the organizations it reaches, and Revoke ends its access immediately — including the access token it is holding. Removing the PostRite connector inside the AI client revokes it too. Posts, templates and media an assistant already created are not affected. Access tokens are also short-lived by design, so even without an explicit revocation an assistant cannot retain access indefinitely.

Common problems

My organization is not on the consent screen

Only organizations whose owner turned on Remote MCP access can be selected. Ask the owner to enable it under Settings → Organization → AI assistants, then connect again.

“MCP access is disabled for this organization”

The owner turned remote MCP access off after you connected. Existing connections stop working until it is turned back on.

The assistant cannot delete a post, template or file

Allow deleting via remote MCP is off — it is off by default. An owner can turn it on in the same settings page.

The assistant says an account needs reconnecting

The social network revoked or expired that account's authorization. Reconnect it in PostRite; posts targeting it are refused until then.

A post fails for a missing option

Pinterest needs a board, a Discord bot needs a channel, TikTok needs a privacy level and WhatsApp a recipient. Ask the assistant to look them up — it can list your boards, the bot's channels and your TikTok account's options.

No best times or suggested times

Insights come only from your own published posts with collected metrics, so a new account, or one on a network PostRite does not collect analytics for, has none yet. The assistant says so rather than inventing a time.