MCP connector

ReBrandOS is a remote MCP server. Connect it to an AI client and the client can read your brand, save drafts it wrote, render carousels and short videos from your own media, and schedule posts — but only after you approve the exact plan. ReBrandOS does not write copy or generate images for you: your AI client writes, and ReBrandOS checks, stores, renders and publishes.

Endpoints

The server speaks MCP over Streamable HTTP and answers with JSON. There is no separate SSE endpoint.

Connect Claude

1. In Claude, add a custom connector (Claude calls them connectors; which Claude plans can add custom ones is up to Anthropic).
2. Paste https://ops.rebrandos.com/mcp as the URL and save.
3. Choose Connect. ReBrandOS opens in your browser; sign in if asked.
4. The consent page says Connect Claude to your brand workspace and lists the permissions Claude asked for. Under Brand access, keep All current and future brands or pick Choose specific brands and tick the ones Claude may use.
5. Choose Accept. Back in Claude, ask it to list its ReBrandOS tools. If it can, you’re connected.

Connect ChatGPT

ChatGPT connects to ReBrandOS as a custom MCP server with OAuth sign-in. Custom servers are only available where OpenAI enables them for your plan and workspace; if you don’t see the option below, check OpenAI’s help center for your plan.

1. In ChatGPT, open Customize → Plugins, then choose Add → Add custom MCP server.
2. Name it ReBrandOS. Under Connection keep Server URL and enter https://ops.rebrandos.com/mcp. Leave Authentication on OAuth; the advanced OAuth settings are discovered automatically.

3. Read ChatGPT’s warning about custom servers, tick I understand and want to continue, then choose Create as a plugin.

4. ChatGPT asks you to connect. Choose Continue to ReBrandOS.

5. Sign in to ReBrandOS if asked. On the ReBrandOS consent page, choose All current and future brands or pick specific brands, review the permissions, and accept. ChatGPT now lists ReBrandOS under Installed.
6. Start a chat and ask for what you need, for example “Make a 5-slide carousel for my brand about our autumn menu and plan it for Thursday at 6 pm.” Nothing publishes until you approve the plan.

The approve-and-schedule card is built as an MCP App. If your client doesn’t display it, ask the AI to show you the plan (brand, account, time, caption) as text before you approve.

Fixed-key option

Use this only if browser sign-in won’t complete in your client. The key is a long-lived secret, so prefer OAuth where you can.

1. In ReBrandOS open Settings → Agent access keys (owners and admins). Under Connect an AI client, find OAuth not completing? and choose Create fixed key.
2. The key starts from the Content agent permissions and a 30-day expiry. Content agent can plan schedules but not commit them; add publish:execute if the client should schedule after you approve.
3. Choose Create fixed key and copy the value now. It is shown once.
4. In Claude, add a custom connector with the URL https://ops.rebrandos.com/mcp/key, choose No sign in, and add a request header named Authorization with the value Bearer rbos_live_….

The key goes only in the header; ReBrandOS never reads keys from the URL. Rotate or revoke it on the same settings page.

Tools by job

Your client only sees the tools its permissions allow. On the fixed-key endpoint there is one more, whoami, which returns the key’s identity.

Workspace and brand context

Accounts and media

Render carousels and videos (no AI model involved)

Drafts, briefs and feedback

Schedule and approve

Results

Personas

Meta campaigns and chatbot flows (Meta engagement: Growth and Enterprise)

Approve before anything publishes

Every post goes through four steps, and the last one only succeeds with your approval of the exact plan:

1. Validate. validate_draft checks the text and media against the network’s limits and your brand rules. A protected client name blocks the draft; words to avoid are warnings.
2. Save. create_draft stores the draft and returns its digest, a fingerprint of the exact content.
3. Plan. plan_schedule checks the account, time slot and the network’s capacity, then returns a plan with its own digest and an expiry 15 minutes away. Your client shows the approve-and-schedule card: brand, where, when (with time zone), format and the caption.
4. Commit. When you choose Approve and schedule, the card posts your approval into the chat and the AI calls commit_schedule with that plan digest, confirm: true and your approval. Nothing is scheduled if the draft changed, the plan expired, the account changed or capacity ran out — the AI has to plan again and ask you again.

Example conversation

You: Turn these onboarding notes into a LinkedIn post for Northwind Coffee
     and schedule it for Tuesday at 9:00.

AI:  [get_brand_context, validate_draft]
     Here's the post. It passes Northwind's rules. Save it?

You: Yes.

AI:  [create_draft, list_accounts, plan_schedule]
     ┌ Approve and schedule ─────────────────────────┐
     │ Brand   Northwind Coffee                      │
     │ Where   LinkedIn · Northwind Coffee           │
     │ When    Tue 20 Oct, 09:00 (Europe/London)     │
     │ Format  Text post                             │
     │ "Week one at Northwind starts with a cup…"    │
     │ [Approve and schedule]  [Change something]    │
     └───────────────────────────────────────────────┘

You: (choose Approve and schedule)
     I approve schedule plan brandos_agent_schedule_plan_… exactly as shown
     (plan digest 4f1c…). Commit it now.

AI:  [commit_schedule]
     Scheduled for Tue 20 Oct, 09:00. That's 1 of today's 10.

What commit_schedule sends

{
  "planId": "brandos_agent_schedule_plan_…",
  "planDigest": "4f1c…",
  "confirm": true,
  "approval": {
    "approvedBy": "you@northwind.example",
    "approvedAt": "2026-10-19T16:05:00Z",
    "approvedDraftDigest": "9a7e…",
    "approvedPlanDigest": "4f1c…",
    "approvalChannel": "claude",
    "approvalSummary": "LinkedIn, Tue 20 Oct 09:00"
  }
}

approvedBy must be the person who connected the client (the identity get_profile returns), and approvedAt must fall inside the plan’s lifetime. For several posts, plan_schedule_batch returns one batchDigest for up to 20 plans, each with its own account and time, and commit_schedule_batch commits them only if none changed.

See Approve and schedule from chat for the card itself.

Automations: comments and DMs

Ask your AI for an engagement automation in plain words: “When someone comments BRAND on this post, DM them the link.” It calls get_automation_guidance first, which says whether your plan includes Meta engagement, which of your Facebook Pages and Instagram accounts can run automations, and whether each one granted the messaging permission. Ask “what automations can you build for me?” and it answers from the same list.

Which one to pick

A post and its automation in one conversation

1. get_automation_guidance { brandId } — check plan, accounts, permissions and webhook health.
2. create_draft — the post copy (e.g. "Comment BRAND if you’re interested").
3. plan_schedule (once per account: Facebook Page and/or Instagram) → show the user → commit_schedule with approval.
4. validate_meta_campaign { brandId, accountId, candidate } — targetMode "rebrandos-draft", targetDraftIds [draftId], keyword, DM text. Fix every error using its fix hint.
5. save_meta_campaign_draft — returns campaign.campaignDigest. One campaign per account (Facebook and Instagram are separate).
6. Show the user who it answers, the keyword, the exact DM and public reply text, and the campaignDigest. Ask for approval.
7. activate_meta_campaign { confirm: true, approval: { approvedBy, approvedAt, approvedCampaignDigest } }.
8. get_meta_campaign — draftTargets shows pending until the post is live, then resolved with the provider post id.

You: Create a post for my Gideon Awolesi brand that says "Comment BRAND if
     you're interested", schedule it for Friday 10:00 on Facebook and Instagram,
     and when someone comments BRAND, DM them: "Here's the brand kit: …"

AI:  [get_automation_guidance, create_draft, plan_schedule ×2]
     Here is the post and the plan for both accounts. Approve?
You: Yes.
AI:  [commit_schedule ×2, validate_meta_campaign ×2, save_meta_campaign_draft ×2]
     Two automations, one per account, both waiting for the post to go live:
     comment contains "BRAND" → one private DM, then reply "Sent! Check your DMs 📩".
     Approve turning them on? (digests 7c2e…, 91ab…)
You: Yes.
AI:  [activate_meta_campaign ×2]
     On. They start answering as soon as Friday's post is live.

What activate_meta_campaign sends

{
  "brandId": "brandos_brand_gideonawolesi",
  "accountId": "sacct_…",
  "campaignId": "camp_…",
  "operationKey": "activate-brand-dm-001",
  "expectedRevision": 1,
  "confirm": true,
  "approval": {
    "approvedBy": "you@example.com",
    "approvedAt": "2026-10-16T09:12:00Z",
    "approvedCampaignDigest": "7c2e…",
    "approvalChannel": "claude"
  }
}

The campaign your AI saves

{
  "contractVersion": 2,
  "name": "BRAND → DM + reply",
  "objective": "comment_autoreply",
  "triggerType": "keyword_match",
  "triggerKeywords": [
    "BRAND"
  ],
  "targetMode": "rebrandos-draft",
  "targetDraftIds": [
    "<draftId from create_draft>"
  ],
  "eventWindow": {
    "timezone": "<the user’s IANA time zone>",
    "start": "00:00",
    "end": "23:59",
    "weekdays": [
      0,
      1,
      2,
      3,
      4,
      5,
      6
    ]
  },
  "recurrence": {
    "mode": "event"
  },
  "replyTemplate": "Sent! Check your DMs 📩",
  "templateSnapshot": {
    "mode": "snapshot",
    "content": "Sent! Check your DMs 📩"
  },
  "metadata": {
    "enableCommentReply": true,
    "enablePrivateReply": true,
    "privateReply": {
      "version": 1,
      "message": "<the exact DM text>"
    }
  }
}

targetMode: "rebrandos-draft" ties the campaign to the post published from that draft on that one account, so it can be set up before the post exists. Until the post is live it answers nothing, and it never answers your other posts. get_meta_campaign shows each draft as pending and then resolved with the network’s post id.

Meta’s rules

• Meta allows exactly ONE private reply (DM) per comment. ReBrandOS records a durable claim per comment, so two campaigns, a redelivered webhook, a retry or a restart can never DM the same comment twice.
• A private reply must be sent within 7 days of the comment (Instagram Live comments: only while the broadcast is live). Later attempts fail with meta_effect_window_expired and are not sent.
• After the private reply, you can only message the person again if they reply; then the normal 24-hour messaging window applies and DM flows (chatbots) take over the conversation.
• The DM text is at most 1000 bytes of UTF-8 (Instagram's limit), plain text.
• Comments by the Page/account itself never trigger automations, so a public reply containing the keyword cannot loop.
• If Meta does not confirm a send (timeout, lost response) the effect is marked ambiguous and is never retried automatically; the public "check your DMs" reply is not sent after an unconfirmed DM.
• All sends go through the shared Meta call governor: Meta rate-limit errors (codes 4, 17, 32, 613, 80000–80014) pause sends and retry later with backoff (up to 4 attempts). Plan for bursts on viral posts being delayed, not dropped.
• Automations answer people who engaged with the account. Never use them for unsolicited outreach, and keep DM content relevant to what the person asked for.
• Private replies need Advanced Access to pages_messaging (Facebook) or instagram_manage_messages (Instagram) on the ReBrandOS Meta app; until then they only reach people with a role on the app.

Permissions

The consent page groups what the client asked for. These are the permissions and the names it shows:

Access tokens last one hour and your client refreshes them; refresh tokens last 30 days and change on every use. Every call is checked against your current role and brand access, so removing someone from a brand takes effect immediately.

Limits and fair use

• 300 requests a minute per AI client, per person, per workspace. Over that, the server answers HTTP 429 with a Retry-After header.
• 10 scheduled posts a day per AI client, per person, per workspace, counted when a post is scheduled and reset at 00:00 UTC. Over that, the tool returns publish_cap_reached.
• Disconnecting and reconnecting the same client does not reset either limit.
• Each network also applies its own posting limits, which ReBrandOS cannot raise. See the fair-use section of the Terms.

When your plan changes

Your subscription state decides what the connector may change:

Writes include saving drafts, uploading and rendering media, planning, committing or moving scheduled posts, and changing Meta automations. A read-only workspace can always cancel queued posts (cancel_schedule), pause or archive Meta automations, read and download its media, and export its data from the dashboard.

When a tool fails

A failed tool call returns isError: true with a JSON body such as {"ok": false, "code": "schedule_plan_expired", "error": "…"}. The codes you will meet most:

Disconnect or pause

Remove the connector in your client, or revoke the connection in Settings → Agent access keys, where OAuth connections and keys are listed together. Pause all agent access on the same page stops every connector, key and webhook in the workspace until you turn it back off.