ViewPrinter のリモート MCP サーバーとは何ですか?
ViewPrinter のリモート MCP サーバーは、https://viewprinter.tech/api/mcp にホストされた Model Context Protocol エンドポイントです。AI アシスタントはローカルに何もインストール・実行することなく接続できます。
ViewPrinter MCP Server認証済み注目
@VMGM Software
基本情報
ViewPrinter MCP Server について
AI アシスタントから TikTok、Instagram、Facebook、YouTube、X へのソーシャル投稿を予約・公開します。アカウントを一度接続したら、何をいつ投稿するかを伝えるだけ——ViewPrinter がキューに入れて公開し、各プラットフォームで何が起きたかを個別に報告します。アカウントごとに OAuth、貼り付ける API キーは不要です。
接続情報
https://viewprinter.tech/api/mcpセットアップ
claude mcp add viewprinter --transport http https://viewprinter.tech/api/mcpViewPrinter MCP Server のツール
23What each platform accepts. Called with nothing, it lists every platform with the media kinds and counts it takes and its caption limit — enough to decide where a post can go. Name platforms to add their recommended sizes and durations, their options with what each one does, and the quirks that decide how a post is treated. Read this before scheduling rather than discovering the rules by being rejected.
List every social account you can post to, across all your workspaces. Each account includes the workspace it belongs to and any groups it is in. Use the returned ids to name post destinations explicitly, or a group name to name several at once. metrics true adds how each account is performing — followers, posts, total likes and following, each with how far it moved over the period — and orders by followers, largest first. Numbers are as of the last sweep rather than live; every account carries the time it was measured, and one nobody has measured yet has no metrics at all rather than zeroes.
Connect a social account. Opens the platform in the person's browser so they can sign in and approve access — that part cannot happen here, because it needs their session and their consent. Call it again after they finish and it reports the connected account. With `invite`, the link is for someone else instead — a friend or family member who will create a brand-new account and connect it, with no login of their own here. They get a page showing the profile to set up (picture, username, name, bio) with copy buttons, the steps to switch Instagram to a Creator account, and a Connect button. Give the person the `url`. Calling again with the same platform and username returns the same invite, updated with whatever profile fields are given, and its status — so that is also how to check whether they have finished. Upload the profile picture first with media_upload, and pass its id once the PUT succeeds.
The products your accounts post for — an app, a book, a shop — each with the accounts linked to it and the workspace it is in. An account belongs to at most one product. Use products_save to create one or change which accounts belong to it.
Create a product, or change one. Found by id when given, otherwise by name (ignoring case) and created when there is none, so saving the same product twice changes nothing the second time. account_ids is the complete list of accounts that post for it: accounts in another product move here, and accounts in this one that are not listed leave it. Leave it out to keep the accounts as they are. A product holds accounts from one workspace; when you are in several and name no accounts, pass workspace_id (from accounts_list).
Delete a product. Its accounts stay connected and belong to no product afterwards. Refused while it has campaigns. Its knowledge documents and gathered research are deleted with it, so this is refused while it has any unless confirm_discard is true — read the refusal, which says how many would go, before confirming.
Your named sets of accounts. Each group lists its members with the state of each, and how many of them need reconnecting — a post naming a group is refused outright if any member cannot publish, so check here before composing one. Use a group name with posts_save instead of listing account ids.
Create a group or change one. A group is a name for a set of accounts, so posts_save can take the name instead of every id. Found by name without regard to case: an existing group is updated (new_name renames it), and a new name creates one, which needs account_ids. account_ids is always the full membership afterwards, not a list to add. All the accounts must be in one workspace.
Remove a group. The accounts in it are not touched and stay connected — only the name and its membership go. Posts already scheduled through it are unaffected, because a post records the accounts rather than the group.
Get a file into ViewPrinter, under an id you can pass to posts_save, media_save or anything else that takes a media id. If the person attached the file in this chat, or has it on their phone or computer, call this with no arguments, as an ordinary tool call — not from code, and without trying to send the file yourself. Apps that show widgets put an upload box under the call, which sends the file to ViewPrinter directly even when you have no network access: tell the person to use it. It takes one video or up to ten photos and posts their media ids to the chat, in order, when they are in. Where no box appears, give the person `upload_page`: any device, no sign-in, lasts an hour. Only if you hold the file yourself and can make HTTP requests (a terminal agent, a server), pass kind and mime_type and PUT the bytes to `url` with exactly the `headers` returned. Apps that show widgets never get `url`: the upload box is the way there. Add sha256 (hex) and a file this workspace already has comes back as existing: true, with nothing to upload. If the file is already on the web — a model’s output link, a meme’s direct link — pass from_url instead, with any description, tags and metadata: ViewPrinter fetches it itself and answers fetching: true straight away. It is recorded within minutes; check with media_list and media_id, and if it never appears the link could not be fetched. purpose: source keeps raw material — a generated take, a recording, a sound effect, a saved meme — to find and reuse later. media_list shows it only when asked for purpose source, and it is never posted or used by a campaign as it is. Several files at once: items, up to 50, each as you would pass one; the answers come back in items, in order. Either way the file lands under `id` and is recorded as soon as it arrives. With the box or the page, wait for the person before using it.
Files in your workspaces, newest first, 50 at a time. Use these ids when attaching media to a post. By default these are the files to post; purpose: source lists the raw material kept to reuse instead. Narrow with kind, tags (files carrying every one) or search (words in the description or tags), or pass media_id for one file of either purpose. Each file has a url to download it from. When nextCursor is not null, pass it as cursor for the next page.
Say what a file is, in your own words. description is what it shows: on a base image, who is on camera — "a woman in her early twenties, blonde hair in a loose bun" — and a campaign reads it so the same person appears in every render instead of changing between them. tags are labels to filter by, like sound-effect; metadata is how it was made — provider, model, prompt, jobId, costUsd, sourceUrl, ref, inputs. Each replaces what is there: null clears a description or metadata, [] clears tags. purpose switches a file between library, to post, and source, raw material kept to reuse. For several files at once, items, up to 50: a file not recorded yet comes back in not_found, to save once media_list shows it. Nothing else about a file can be changed here; what it is was read from the file itself.
Remove a file and its stored bytes. This cannot be undone. A file a post still uses cannot be removed, and neither can raw material kept with purpose source: switch it to library with media_save first.
Comments on your Facebook and Instagram posts, newest first, replies included. Pass target_id — the id of one destination in a posts_list post's targets — for that delivery's thread, or account_id for everything on one account. Threads: externalParentId names the externalCommentId of the comment it answers; null is top-level. authorKind says who wrote it, as far as the platform would say: "account" is the connected account itself (your own reply), "person" a named member of the public, and "withheld" means the platform would not say — Facebook withholds commenters for now. Withheld is not anonymous; do not report it as such. Read hourly for a post's first day, daily to a week, weekly to thirty days; lastSeenAt is when each was last confirmed. A comment its author deleted has goneAt set and is left out unless include_gone is true. When nextCursor is not null there are more: pass it as cursor with the same arguments for the next page.
Reply to a comment on one of your posts. The reply is posted publicly, immediately, in that account's name, and cannot be unsent from here. comment_id is a comments_list row's id, not the platform's. The reply is threaded under that comment. It returns status "sent" only when the platform gave the reply an id. "pending" means the platform did not answer and the outcome is unknown: it is confirmed within the hour, and retrying with the same idempotency_key is safe and will not post twice.
Direct messages sent to your Instagram and Facebook accounts, newest conversation first. Call with no argument for the inbox, account_id for one account, or conversation_id for one thread and its messages. One conversation is one person on one of your accounts: the platforms carry no thread id, so the person is the thread. replyUntil is the deadline for answering and it is the point of this tool. Meta allows a reply within 24 hours of their last message and refuses after it, so a null replyUntil means a reply will be rejected however good it is — say so rather than drafting one. closesInSeconds counts down to the same moment. participant may be null. The webhook carries ids and no names, so an unnamed person is the platform not saying, not an anonymous one — do not report them as anonymous. fromUs marks our own messages, which include replies sent from the platform's own app. A message the sender unsent has gone set and keeps its place, so a reply still reads in order. The inbox comes 50 conversations at a time; when nextCursor is not null, pass it as cursor with the same arguments for more.
Send a direct message, privately, in the account's name. Name exactly one of conversation_id or comment_id. conversation_id answers a thread from messages_list, and only while its replyUntil has not passed — Meta allows a reply within 24 hours of their last message and refuses after it. comment_id opens a thread from a comment in comments_list: somebody who comments on your post can be messaged once, within 7 days. One message, not a conversation — they have to answer before you may send another. Use it when a comment needs an answer that does not belong in public. You can never message someone first. If they have neither written nor commented, there is no way to reach them and no permission that grants one. This is private and comments_reply is public. Answering in the wrong one either exposes something meant for one person or hides an answer everyone was waiting for.
The webhooks set up in your workspaces: where each one delivers, which events, any product or account filter, and whether it is working. status "paused" means too many deliveries in a row failed; lastError says why, and webhooks_save with its id resumes it. hasAuthHeader says whether a credential is set, never what it is.
Ask to be told when something happens: each matching event is POSTed as JSON to your https URL within seconds. Create one, or change one by id — saving also resumes a paused webhook. Events are thin: a type, a stable id, the workspace, product and account, and the ids to look things up by — never the comment or message text. On comment.created, read it with comments_list comment_id and answer with comments_reply or messages_reply comment_id before privateReplyUntil. On message.received, read the thread with messages_list conversation_id and answer before replyUntil. Answer 2xx within 10 seconds. A failed delivery is retried three times over about twenty minutes, so the same event id can arrive more than once: ignore ids you have already handled. Only comments and messages from the last day are announced. auth_header is sent as the Authorization header on every delivery, for an endpoint that expects a credential (e.g. "Bearer …"). It is stored encrypted and never shown again; null removes it.
Stop a webhook. Events already on their way may still arrive.
Ask a question about your niche and get back the posts that answer it, from TikTok and Instagram unless platforms narrows it. Give question and product_id and it starts searching for that question, which takes a few minutes, and returns at once with a thread_id and status working. Call it again with just the thread_id, about every 15 seconds, until status is ready; then posts are the answer, best first. Each has its URL, caption, hashtags, views, likes, comments, shares and saves, the creator’s follower count, and outperformance: views as a multiple of that creator’s median over the posts we hold of theirs, null when we hold too few. fit says how well a post suits your product (in_niche, adjacent, off_topic) and posts judged off-topic are left out unless you ask for all. A thread takes one question at a time: asking again while it is working does not start a new one. With a failed status, ask the question again. Starting a question uses search credit, so do not repeat one that is working. With neither question nor thread_id it reads what is already indexed, instantly. A post we have watched also carries hook, genre, format and repeatableStructure (null until watched), and a question’s answer gives each post a why: one line on what the video is and how it was judged. The top-level formats and ideas are empty for now.
Create, change or cancel a post. This publishes: a new post is queued for every account named and goes out at scheduled_at, or as soon as possible without one. Do not tell anyone their post is saved or scheduled until this has returned successfully. No post_id creates a post. Name accounts with ids from accounts_list, or a group from groups_list for all of its members — the two combine, and a group is expanded now, so adding an account to it later does not add it to this post. Accounts must all be in one workspace. Attach media by id from media_list; a post is either a video or a set of images, never both. draft true holds it here without sending anything. With post_id, changes a post that has not gone out: caption, time, destinations, or draft state. account_ids and groups then replace the destination list entirely. draft false sends a draft. Setting scheduled_at on a post whose destinations failed queues the failed ones again and leaves the published ones alone — fix what made them fail first, usually by reconnecting. The caption and destinations can only change while nothing has published; after that only the time can, and new words mean a new post. With post_id and cancel true, stops it: destinations that have not started are cancelled, and any already being published cannot be recalled and come back in still_going. Safe to repeat. Publishing has to be enabled on the account; where it is not, a new post saves nothing at all and a draft is left as it was, and the answer says so. Cancelling works whether or not publishing is enabled.
Posts in your workspaces, newest first, one page at a time, each with the state of every destination it was sent to and the numbers it has earned so far. Call it to find one post (post_id), to check whether something published, and before changing or cancelling anything — posts_save takes a post_id from here. Numbers are as of the last sweep, and every destination carries the time it was measured — one nobody has measured yet has no metrics at all rather than zeroes. `trial` names an Instagram trial reel's graduation strategy, and is null for anything that is not one. A trial is shown to non-followers first, so its numbers come from a colder audience than an ordinary post's: compare trials with trials. Each row also carries `insights`: what the platform's own analytics said on that delivery, never summed across deliveries. reach (distinct accounts), totalWatchMs and avgWatchMs, replays, follows (gained from the post), completionRate and skipRate as fractions 0..1, `retention` — the full audience-retention curve as Facebook sends it, `points[i]` being the share still watching at i twentieths of the video and `bucketMs` one twentieth in milliseconds — and `breakdowns` such as clicks by type. retentionAt3s and retentionAtHalf are read off that curve, interpolated between buckets, so hooks can be ranked by how many viewers they held; null without a curve or a known video length. Null means the platform does not report it; a name listed in `unavailable` means it would have and refused this time. Compare watch time across platforms as totalWatchMs / views: Facebook averages per play and Instagram per viewer, so avgWatchMs is not comparable between them. Every filter is optional and they combine: a date range, statuses, platforms, accounts, groups and products by name, where the post came from, and a caption search. Pass nextCursor back as cursor for the next page; null means there is no more.
概要
What it does
ViewPrinter publishes to TikTok, Instagram, Facebook, YouTube and X. Connect an account once in the browser, then drive everything from your assistant: upload media, queue a post to one account or a named group, change it while it is still pending, cancel it, and read how it did.
It does not write your content. No captions, no images, no video. Your agent's model is better at that than any feature here would be — ViewPrinter is the part that actually gets it posted.
Connecting
OAuth only. There is no API key, and nothing to paste into a config file.
{
"mcpServers": {
"viewprinter": { "type": "http", "url": "https://viewprinter.tech/api/mcp" }
}
}
For Claude Code: claude mcp add -t http viewprinter https://viewprinter.tech/api/mcp
The authorization server uses Client ID Metadata Documents rather than Dynamic Client Registration, which MCP 2026-07-28 deprecates. A client that only implements DCR will say the auth server is incompatible; point it at a metadata document instead. One is served at https://viewprinter.tech/.well-known/openclaw-client.json for clients that do not publish their own.
The 18 tools
| Area | Tools |
|---|---|
| Platforms | platforms_list — each platform's media rules, caption limits and options |
| Accounts | accounts_list, accounts_connect, accounts_performance |
| Groups | groups_list, groups_create, groups_update, groups_delete |
| Media | media_upload, media_confirm, media_list, media_update, media_delete |
| Posts | posts_schedule, posts_list, posts_update, posts_cancel, posts_performance |
Every tool declares its annotations explicitly — read-only, destructive and open-world are all set on all 18, so a client knows what it is about to do before it does it.
Worth knowing
- A post is either a video or a set of images, never both. YouTube needs a video; X is the only platform that takes a caption with no media.
platforms_listis the source of truth — call it before composing rather than discovering the rules by being rejected. - Media upload is two steps.
media_uploadreturns a short-lived URL to PUT the file to; the bytes go straight to storage and never through the conversation. Nothing is recorded untilmedia_confirm. - Groups are expanded when you schedule, not linked. They are shorthand for the accounts in them at that moment.
- Metrics come from a background sweep, and every row carries the time it was measured. Something nobody has measured yet has no metrics rather than zeroes.
Links
- viewprinter.tech
- Setup for other clients
- Claude Code plugin
- Listed in the official MCP registry as
tech.viewprinter/viewprinter
よくある質問
ViewPrinter の MCP サーバーにはどう接続しますか?
エンドポイント https://viewprinter.tech/api/mcp を Claude Code、Cursor、VS Code など任意の MCP 対応クライアントに追加してください。このページのセットアップ手順で各クライアントを一度に設定できます。
ViewPrinter の MCP サーバーは認証が必要ですか?
はい。ViewPrinter は OAuth を使用します。初回接続時に MCP クライアントがブラウザを開いてサインイン・認可を行い、以降はその認証情報を再利用します。
ViewPrinter の MCP サーバーはどのトランスポートを使用しますか?
ViewPrinter は Streamable HTTP エンドポイントを公開しており、これはリモート MCP サーバーで広く使われ、主要な MCP クライアントがサポートするトランスポートです。
他のリモート MCP サーバー
42min
Official 42min MCP server: let Claude, ChatGPT, Cursor or any MCP client manage your meeting scheduling. Check availability, find free slots, book, cancel and reschedule meetings, manage recurring series, secret booking
A
AIS Gateway
CVE prioritization using exploitation signals, plus agent heartbeat monitoring and state capsules. Some tools use API-key or x402 access.
9apes
The 9apes MCP server lets an AI assistant read the GitHub Actions data 9apes already collects for your organization: runs, jobs, steps, and their durations. Ask it why a build got slower, which jobs take the longest, or
AdWhispr
Research competitors' winning ads, clone them for your brand and launch paused campaigns on Meta, Google and TikTok from chat.
ARVX
Email for long-term client relationships: ARVX reads each person first and remembers every pair.
コメント