Genviral

$ claude mcp add --transport http genviral https://mcp.genviral.io/mcp

Adds Genviral to the current project. Run /mcp afterwards to complete the OAuth sign-in.

Genviral publishes to

TikTokInstagramYouTubePinterestLinkedInFacebookXBlueskyMastodonTelegram

In practice

Things people actually type

Each one resolves to real tool calls against your account — reading what is already scheduled, generating what is missing, and queueing the result.

• What did I post last week, and which one did best?
• Generate three TikTok slideshows about morning routines and schedule them for this week.
• Move everything scheduled for Friday to Monday morning.
• Which of my connected accounts hasn't posted in the last 10 days?
• Make a 9:16 video from this image and queue it as an Instagram Reel.
• Draft a LinkedIn post from this changelog, but don't publish it.

Two layers

Your assistant decides. Genviral executes.

The assistant is good at knowing what to make and when it should go out. Genviral is good at being the one connection that can actually make it and publish it.

Your assistant owns

• It already knows what you meant You describe the week you want. The assistant picks the format, writes the captions, and decides what goes where — no command syntax, no dashboard.
• It runs where you already work Claude, ChatGPT, Cursor, VS Code, and every other MCP client read the same server. Connect once per client, not once per feature.
• It carries your context Whatever else is in the conversation — a launch doc, a repo, a spreadsheet of hooks — becomes source material for the posts it drafts.
• It closes the loop Analytics come back through the same connection, so the next batch is written against what actually performed instead of what you hoped would.

Genviral owns

• Connected social accounts 10 platforms behind one OAuth connection — no Meta or TikTok partner approval wait, and no per-platform tokens in your assistant's config.
• AI media generation Slideshows, images, and video rendered in Genviral's Studio. The assistant asks for an asset and gets a finished one back — most MCP servers can only schedule what you already made.
• Duplicate-safe publishing Every write through the MCP server requires an idempotency key, and publishing requires an explicit decision. A retrying agent cannot post the same thing twice.
• Workspaces and teams Each connection is bound to the personal space or one workspace you pick during consent, so an agency assistant cannot reach a client it was not pointed at.

Connect

One URL, in under a minute

There is no package to install and no server to run. Pick your client, paste the endpoint, and approve the permissions you want it to have.

$ claude mcp add --transport http genviral https://mcp.genviral.io/mcp

Adds Genviral to the current project. Run /mcp afterwards to complete the OAuth sign-in.

1. 02Step 2 of 4

   Add the server to your client

   One remote URL, nothing to install and no process to keep running. Claude Code takes it as a command; Cursor and VS Code take it as four lines of JSON; Claude and ChatGPT take it in their connector settings. Claude Code $ claude mcp add --transport http genviral https://mcp.genviral.io/mcp
2. 03Step 3 of 4

   Approve the permissions you actually want

   Signing in opens a Genviral consent screen. Pick the workspace the connection is bound to and grant only the scopes you want it to have — reading and drafting are separate from publishing, and generation is separate again. Revoke any connection later without touching the others. Create a free account
3. 04Step 4 of 4

   Ask for one post and watch it land

   Start with a draft rather than a publish, review the caption and media in the dashboard, then let the assistant schedule the rest. Media it generates comes back as a gallery in clients that render MCP resources. In your assistant $ Draft one TikTok slideshow about this week's launch, $ put it in my content library, and stop before publishing.

Tools

32 tools, 4 permissions

This list is generated from the same contract the server authorizes against, so it cannot advertise a tool the server does not register. Everything here executes real work — there are no stubs that hand you a link and ask you to finish the job.

Read your account

context:read

Accounts, posts, analytics, and library contents. Nothing on this list changes anything.

• show_studio Read only Show finished Studio images and videos once they are ready.
• show_posts Read only Show your content calendar, or one post in it, when you ask to see it.
• show_video Read only Show a video project with a playable preview, and open the full editor from it.
• show_library Read only Show a visual preview of your files or slideshows when requested.
• show_viral_library Read only Show viral reference cards when you want to browse inspiration.
• show_analytics Read only Show an analytics dashboard for your posts or one tracked account when you ask to see it.
• browse_music Read only Browse the music library and your own uploaded tracks.
• get_context Read only Connected accounts, workspaces, plan and credit balance, and the Studio models available to you.
• list_posts Read only Scheduled, drafted, and published posts, with per-account delivery status on each one.
• get_analytics Read only Post and account performance, plus every account you track for analytics.
• research_trends Read only Trend briefs for a niche, and a preview of any public TikTok slideshow before importing it.
• research_reddit Read only Search Reddit for source posts, then read bounded comment samples from the threads you choose.
• analyze_content Read only Score a hook against real viral posts, suggest proven hook structures, and find viral references for an idea.
• search_viral_library Read only Search Genviral's curated library of viral posts, and open any one of them in full.
• browse_library Read only Files, folders, influencers (consistent characters), image packs, templates, and slideshows in your content library.
• search_images Read only
• get_job Read only Status of a running render, generation, or analytics refresh.

Create and edit content

content:write

Drafts, uploads, folders, packs, templates, and slideshows. Still cannot publish.

• manage_music WritesCan remove data Add, rename, import, and delete your own music tracks.
• manage_studio WritesCan remove data Save Studio creations into your library, delete them, or favorite a template.
• upload_media Writes Upload an image or video into your media library.
• manage_library WritesCan remove data Create, move, and delete folders, and move items between them.
• manage_assets WritesCan remove data Create and edit image packs and reusable slideshow templates.
• manage_slideshow WritesCan remove data Duplicate, edit, render, or delete a slideshow, or regenerate a single slide.
• manage_analytics_targets WritesCan remove data Add, edit, remove, and refresh the accounts you track for analytics.
• create_post Writes Prepare a post: media, caption, per-platform settings, target accounts, and schedule.
• update_post WritesCan remove data Reschedule, edit, retry, or delete posts you already created.

Schedule and publish

posts:publish

The only scope that can put something in front of an audience. Grant it separately.

• schedule_editor_versions Writes Prepare and schedule rendered versions of a video project across your accounts.
• publish_post WritesCan remove data Commit a prepared post — publish it now or hand it to the scheduler.

Generate media

media:generate

Image, video, and slideshow generation. This is the scope that spends credits.

• edit_video WritesCan remove data Create and edit video projects: clips, text, captions, and music, then render the final cut.
• remix_video Writes Find source videos to remix, preview them, and import clips into a project.
• generate_slideshow WritesCan remove data Generate a multi-slide carousel from a brief, or import one from TikTok, then render it.
• generate_studio_media WritesCan remove data Generate images and videos in AI Studio from a prompt or a reference image.

Safety

What it can do is your decision, not the model's

Handing an assistant your social accounts is only reasonable if the limits are enforced somewhere it cannot reach. All four of these live in the backend.

OAuth, not a pasted API key

You sign in to Genviral to connect. No long-lived key sitting in a config file, in a shell profile, or in a URL your assistant logs. Revoke a connection and it stops working immediately.

Permissions you choose per connection

4 separate scopes, approved on a Genviral consent screen. An assistant can be allowed to draft and never allowed to publish, or to read analytics and nothing else.

Scoped to one workspace

You pick the personal space or a specific workspace while approving. The connection cannot see, post to, or spend credits from anything outside it.

Publishing is always deliberate

Publishing takes an explicit publish-or-schedule decision, and every write carries an idempotency key. An agent that retries a failed call does not double-post.

Use cases

Six workflows worth connecting it for

Each one is a single request in the assistant you already have open. The loop is the same every time: it reads the account, generates what's missing, and queues the result.

• TikTok A week of slideshows from one sentence The assistant pulls a trend brief, generates five carousels off it, and queues one per day — you approve the batch instead of writing it.
• Instagram Reels captioned from an approved folder Point it at a media library folder. It captions each clip, adds hashtags, and queues them without anything leaving your approval pipeline.
• LinkedIn Posts written from work you already did Because the assistant can see your repo, docs, or changelog in the same conversation, the post is drafted from real detail instead of a summary of a summary.
• YouTube Shorts published on a review gate It uploads, fills the metadata, and publishes unlisted first. You flip it public once you've watched it — the assistant never gets that decision by default.
• Pinterest Infographic pins generated on a cadence Studio renders the image, the assistant writes the pin copy, and the whole thing goes to the right board — one prompt, no design step.
• Facebook Cross-posting without a second workflow One connection covers every platform, so mirroring today's TikTok onto a Facebook page an hour later is a sentence, not an integration.

Guides

What to read next

The developer surfaces behind the connection. Every MCP tool call ends up at one of these endpoints.

• Genviral Partner API reference Every MCP tool is a thin wrapper over these endpoints. Read this when you want the exact request and response shapes behind a tool call. Read guide
• Genviral for AI agents The short, machine-readable overview of both connection paths — MCP for assistants, REST for code — with the same capability list. Read guide
• Genviral CLI The alternative for coding agents: one binary, one scoped API key, and no MCP client involved. Useful for Codex, CI jobs, and cron. Read guide

FAQ

MCP questions

What teams ask before pointing an assistant at their social accounts.

What is MCP?

The Model Context Protocol is an open standard, introduced by Anthropic, for connecting AI assistants to outside tools and data. An MCP server publishes a set of tools; any MCP client — Claude, ChatGPT, Cursor, VS Code, and many others — can call them. It means Genviral only has to expose its capabilities once for every assistant to use them.

What is the Genviral MCP server URL?

https://mcp.genviral.io/mcp. It is a remote streamable-HTTP server, so there is nothing to install and no local process to keep alive. Clients that support discovery can also read https://www.genviral.io/.well-known/mcp/server-card.json to find the endpoint, the full tool list, and the available scopes automatically.

How do I authenticate?

With OAuth. When your client first connects, it sends you to a Genviral consent screen where you sign in, choose the workspace the connection is bound to, and approve the permissions you want it to have. There is no API key to generate, copy, or paste into a config file, and you can revoke any single connection later without affecting the others.

Which AI clients work with it?

Any MCP client that supports remote servers with OAuth. That covers Claude (desktop, web, mobile, and Claude Code), ChatGPT's custom connectors, Cursor, VS Code with Copilot, Windsurf, Cline, Zed, JetBrains IDEs, and anything built on an agent framework with MCP support.

What can the assistant actually do?

32 tools, covering the whole loop: read your accounts and analytics, browse and manage your content library, generate images, videos, and slideshows in Studio, prepare posts, and schedule or publish them to 10 platforms. Generation is the part most social MCP servers do not have — they can schedule content, but they cannot create it.

Can it publish without asking me?

Only if you grant it the publish permission, and only to the accounts in the workspace you connected. Reading, drafting, and generating are separate scopes, so a connection can be allowed to build a whole content queue while remaining unable to put any of it live. Publishing also requires an explicit publish-or-schedule decision on every call rather than defaulting to going out immediately.

Will a retrying agent double-post?

No. Every write through the MCP server requires an idempotency key, so a retried call resolves to the same post instead of creating a second one. That guarantee comes from the same backend the dashboard and Partner API use, not from anything the assistant has to remember to do.

Do I need a paid plan?

You can create a free Genviral account and connect it. Generating media spends credits, and connected-account limits follow your plan, so heavier automation needs a paid plan the same way it does in the dashboard.

Should I use MCP or the REST API?

Use MCP when a chat assistant is driving — it is the fastest path from a sentence to a scheduled post. Use the Partner REST API, or the Genviral CLI, when you are writing code, running scheduled jobs, or building a product on top of Genviral. Both hit the same backend services, so nothing is exclusive to one path.