Schedule social posts from ChatGPT

# posterly for ChatGPT

posterly is the publish layer. ChatGPT is the chat layer. Connect the hosted MCP endpoint as a custom app, then schedule across 18 platforms after the user approves.

posterly publishes to 18 platforms: Instagram, Facebook, TikTok, X, LinkedIn, YouTube, Pinterest, Threads, Google Business, Telegram, Bluesky, Discord, Slack, Mastodon, Dev.to, Hashnode, WordPress, Lemmy.

## Connect

ChatGPT web: Developer Mode, create a custom MCP app, server URL https://www.poster.ly/api/mcp. Auth is OAuth or Bearer. Write actions need Business, Enterprise, or Edu. Pro custom MCP is read and fetch only. Web only. No mobile MCP.

Codex: `codex mcp add posterly --env POSTERLY_API_KEY=pst_live_your_key_here -- npx -y posterly-mcp-server@latest` or `codex mcp add posterly --url https://www.poster.ly/api/mcp`. Config: ~/.codex/config.toml.

This is a custom app you add yourself. Do not look for a directory listing.

Stdio file_path uploads work in Codex only, not in ChatGPT web. Chat attachments never reach MCP.

## Are we connected yet?

1. Call `whoami` (no params). Confirm the user, scopes, and workspaces.
2. Call `list_accounts` (`workspace_id` optional). If none are connected, stop and send the user to connect social accounts in posterly. Do not invent account IDs.
3. Prefer `validate_post` before `create_post`. Show the preview. Wait for a clear yes.
4. Only then call `create_post` with `confirm: true`. Never publish or schedule live without that approval.
5. Prefer `scheduled_at` (ISO 8601 UTC) or `find_available_slot` (`account_ids`, `timezone`, `count`). Do not publish immediately unless the user asks.

## Rules

- posterly is the publish layer. This client is the chat or agent layer.
- Never collect card details, posterly passwords, or social-network passwords. Send the user to https://www.poster.ly/agents/signup or https://www.poster.ly/dashboard/api.
- Point to https://www.poster.ly/mcp for the rest of the 86 hosted HTTP MCP tools (89 on the local stdio package).
- After a post is created, send the View in posterly link. Avoid raw JSON dumps unless the user asks to debug.

## Media (honest)

Chat attachments never reach MCP. The paperclip in this chat does not give posterly the file.

The ~4MB limit is the **Vercel request body** on posterly API routes (`/api/mcp`, `POST /api/v1/media/upload`). It is not an MCP protocol cap and not the plan storage cap. Base64 inflates files, so hosted `upload_media` is small images only (relay cap 5MB decoded).

- Public HTTPS URL: `upload_media_from_url` (`url`, optional `filename`, `content_type`) or pass `media_url` / `media_urls` on `validate_post` / `create_post`. Server fetch: about 5MB relay, 60MB hard cap. Large R2/S3 URLs fail with `remote_media_too_large`.
- Hosted `upload_media`: required `filename` + `base64_data`. Optional `content_type`. Hits the 4MB Vercel body. JPEG, PNG, GIF, WebP, MP4, MOV, WebM. HEIC is unsupported.
- Laptop file in ChatGPT web or Claude.ai: `create_media_drop` (optional `filename`, optional `max_files`, default 10). Send the user `drop_url` (`https://www.poster.ly/drop/<token>`). They open it with no dashboard login and drop the file. Then `list_media` (`drop_session_id`) and pass `public_url` to `validate_post` / `create_post`. This is not the chat paperclip.
- Larger files, when this client can HTTP PUT: `create_signed_upload` (`filename`, `content_type`, `size`), PUT the raw bytes to the returned `upload_url` (object storage, not a posterly API route) with the returned headers, then pass `public_url` to `create_post`. That PUT does not hit the 4MB body. Plan video caps: Starter 500MB, Pro 750MB, Power 1GB, Agency 4GB.
- Stdio `upload_media` also accepts `file_path` when the user gives a real local filesystem path. The local server reads the disk and uses signed upload for large files, so this does not hit the Vercel 4MB body. Never guess paths. Chat attachments are not a file_path.

## Starter tools

- `whoami` (none): Call first. Returns the authenticated user, API key scopes, and workspaces.
- `list_accounts` workspace_id?: Connected social accounts. Stop if this is empty and send the user to connect accounts in posterly.
- `validate_post` account_id or username+platform, caption, scheduled_at?, media_url?, media_urls?, post_type?, workspace_id?: Dry run. Does not create a post. Use this before showing a preview.
- `create_post` same as validate_post, plus confirm: true (required): Destructive write. Refuse unless the user clearly approved caption, account, media, and schedule.
- `upload_media` filename, base64_data (hosted). content_type optional. Stdio also accepts file_path.: Hosted path is base64 with a decoded file of 5MB or less. Chat paperclip files never arrive here.
- `upload_media_from_url` url (required), filename?, content_type?: Fetch a public HTTPS URL into posterly storage. Localhost and private IPs are blocked.
- `create_signed_upload` filename, content_type, size: Bypasses the Vercel ~4MB API body. Returns upload_url (object storage) and public_url. PUT raw bytes to upload_url, then pass public_url to create_post. Use this when the client can PUT (Cursor, Claude Code, Codex).
- `create_media_drop` filename?, max_files?: Creates a no-login page at https://www.poster.ly/drop/<token>. Send drop_url to the user. ChatGPT web and Claude.ai laptop files use this, then list_media. Not the chat paperclip.
- `list_media` limit?, drop_session_id?: Newest media assets. After create_media_drop, pass drop_session_id and use public_url on validate_post / create_post.
- `find_available_slot` account_ids?, timezone?, workspace_id?, count?: Always pass timezone explicitly (IANA, e.g. Europe/London). Respects a 1-hour gap.
- `generate_captions` platforms (required), brief?, tone?, count?, mode?, source_caption?: Returns caption options only. It does not schedule or publish.
- `list_posts` status?, platform?, account_id?, workspace_id?, limit?: Upcoming or recent posts. status is scheduled, published, failed, or draft.

## Example prompts

- Using posterly, run whoami, then list_accounts. Tell me which LinkedIn, X, and Instagram accounts are connected.
- Draft one caption for LinkedIn, X, and Instagram about our product launch. Show me the three drafts and wait.
- Find the next free slot in Europe/London for my LinkedIn account, then run validate_post for a text post. Do not create_post yet.
- Turn this blog URL into three scheduled posts (LinkedIn, X, Instagram). validate_post each one, then wait for my yes before create_post with confirm true.
- List my last 10 scheduled posts and summarise what is already on the calendar this week.
- Generate three caption options for LinkedIn with generate_captions, then help me pick one. Do not publish.
- Schedule the approved LinkedIn caption for the next free Europe/London slot. Use create_post with confirm true only after I say yes.
- I have a video on my laptop. Call create_media_drop, send me the /drop/ link, wait, then list_media and validate_post. Do not use the paperclip.

## Links

- Human landing: https://www.poster.ly/chatgpt
- MCP hub: https://www.poster.ly/mcp
- ChatGPT guide: https://www.poster.ly/guides/chatgpt-guide
- AI agents: https://www.poster.ly/agents
- Agent signup: https://www.poster.ly/agents/signup