Model Context Protocol

Your social scheduler, inside Claude

Ravenpost runs an MCP server. Connect it once and an assistant can write, schedule and publish across 9 networks for you — 22 tools over the same services and the same validation the dashboard runs on. No browser automation, no second set of rules.

On every plan, the free one included · sign in to connect, no token to copy · disconnect in a click

22 tools · one conversation

What the assistant can actually do

Everything the dashboard does, minus the clicking. Each card lists the real tool names — they are generated from the server’s own registrations, not written out here.

See where it can post

Your workspaces and every connected account, with platform and status — so the assistant names real destinations instead of guessing at them.

  • list_workspaces
  • list_accounts

Write, schedule, publish

One caption fanned out to any set of accounts, with a different caption per network, threads, replies and quotes. Publish now, schedule a time, or drop it in the queue.

  • list_posts
  • get_post
  • preview_post
  • create_post
  • update_post
  • delete_post
  • schedule_post
  • publish_post

Prepare the media

Upload an image or video, cut one photo into every network’s canvas at once, or hand it a presigned URL so a 200 MB file never travels through the conversation.

  • upload_media
  • create_media_upload
  • attach_media
  • create_variants
  • list_media
  • image_formats

Ask what actually worked

Followers and per-post engagement, plus the best hours to publish measured from your own posts — with an explicit “not enough data yet” rather than a made-up hour.

  • get_analytics
  • best_times

Reel audio

Search Instagram’s licensed audio catalog and attach a track to a reel — the only native music path any of these platforms exposes.

  • list_audio

Pinterest

1 tool in this group.

  • list_boards

Know the rules first

Caption budgets, media counts and posting-queue slots, read from the same registry the publisher enforces — so limits are checked before a post is written, not after it is rejected.

  • list_queue_slots
  • platform_limits

Every tool, with what each one takes and returns: the MCP reference.

Step 1 · Sign in

Connect it like any other app

Add the endpoint, click Connect, sign in to Ravenpost. Your client runs the OAuth flow itself — there is no credential to find, copy or paste anywhere.

OAuth — the way to prefer

The server publishes its authorization metadata, so a client that speaks MCP authorization discovers where to send you and opens the sign-in on its own. You approve the connection once. Disconnecting it in your client ends its access immediately — nothing keeps working until an expiry date.

The only thing to paste
https://api.ravenpo.st/mcp

No browser flow? Use a token instead

Some clients — and every script hitting the REST API — cannot run a sign-in. For those, generate a personal access token on the dashboard’s MCP page. It starts rvp_, is shown once (we keep a SHA-256 hash and a short prefix, so we cannot show it again), carries the workspace it was minted in, and is revocable from the same page.

Step 2 · Connection

Point your client at it

One endpoint, https://api.ravenpo.st/mcp. What differs between clients is only whether they sign you in or want a token.

One command in the terminal

Add the server with no credential at all: the first call answers with a sign-in challenge, and /mcp inside Claude Code walks you through it.

Add the server, then sign in
claude mcp add --transport http ravenpost https://api.ravenpo.st/mcp

claude mcp list

Prefer a token — for a CI box, or an account you don’t want to sign in on? Pass it as a bearer header instead:

With a token
claude mcp add --transport http ravenpost https://api.ravenpo.st/mcp \
  --header "Authorization: Bearer rvp_your_token_here"
Step 3 · In use

Things to ask it

Paste one in once you’re connected. The assistant picks the tools; the server keeps it inside every limit.

Which social accounts do I have connected on Ravenpost?

Draft a post about our new pricing page for Instagram, X and LinkedIn. Give X its own shorter caption and show me all three before you schedule anything.

Take this image, cut it for an Instagram feed post, a story and LinkedIn, and attach the right one to each destination.

What’s my best hour to post on LinkedIn, based on my own numbers?

Schedule this to the next free slot in my posting queue.

Which of last month’s posts got the most engagement, per platform?

Turn this into a five-tweet thread and post the first one as a reply to my launch tweet.

What are the caption limits for the networks I have connected?

Where the limits are

What it will not do

Handing an assistant your accounts is a reasonable thing to hesitate over. These are the answers, and they are enforced on our side rather than asked of the model.

It cannot invent media

The media tools ingest files you provide — a URL, a base64 blob, a presigned upload. Nothing here generates an image or a video, and nothing bills you for generation credits.

The platform rules are the server’s, not the model’s

Caption budgets, media requirements, TikTok’s compliance fields, thread caps. A post that would breach one is rejected with the reason — never silently truncated to fit.

It asks which workspace

With more than one workspace, every tool requires you to name it and the error lists them. Publishing one client’s content to another’s accounts is the failure this design exists to prevent.

A draft stays a draft

Creating a post and publishing it are separate actions. Nothing goes live because a conversation drifted — publishing is a step the assistant has to take on purpose.

Whichever credential you use, you can take it back

Sign in and your client holds a connection you can disconnect, which ends its access at once. Use a token instead and we keep only a SHA-256 hash and a short prefix — it is shown once, carries one user, and revoking it from Settings cuts access immediately.

Same engine as the dashboard

The server builds a fresh session per request over the dashboard’s own services — no second implementation to drift, no shared state between clients or workspaces.

Questions

What is an MCP server, in one sentence?+

Model Context Protocol is the open standard for giving an AI assistant real tools instead of a browser to click around in — so Claude talks to Ravenpost over an API and gets structured answers back, rather than screenshotting a dashboard.

Which networks can the assistant post to?+

All 9 we support: Instagram, TikTok, X, Telegram, LinkedIn, Threads, Bluesky, YouTube, Pinterest. One post fans out to as many connected accounts as you pick, and each destination keeps its own status, permalink and error.

Does it cost extra?+

No. The 22-tool MCP server, the REST API and webhooks are on every plan including the free one. Your plan's limits — connected accounts, posts per month — apply the same way whether a post comes from the dashboard or from a conversation.

Do I have to use Claude?+

No. The endpoint speaks Streamable HTTP and is stateless, so any MCP client works — Claude Code, Claude Desktop, claude.ai custom connectors, Cursor, or something you wrote yourself. Claude is what most people arrive with, not a requirement.

Can it post something without me seeing it first?+

It can, if you ask it to publish — that is the point of the tool. What it cannot do is exceed the limits: a post is created as a draft unless an action says otherwise, platform rules are enforced server-side, and every published post appears in the dashboard with its own status and permalink, so nothing happens off the record.

Do I need to create a token?+

Not for most clients. Paste the endpoint, click Connect and sign in — your client obtains its own credential through OAuth, and disconnecting it there ends access immediately. A personal access token is for the cases a sign-in cannot cover: a client with no browser flow, a script, a CI job, or the REST API.

Is putting the token in a URL safe?+

That form is a fallback for a client that can neither sign in nor send headers. We never log request URLs server-side, and any token can be revoked at once — but where your client offers the sign-in or a header, prefer either of those.

Connect it in about two minutes

Free plan, no card. Generate a token, paste one command, and ask your assistant what you have connected.