Run your social channels from an AI assistant: Klipsy's MCP server

An MCP server lets an AI assistant like Claude or ChatGPT securely call an app's own tools on your behalf. Klipsy ships one: a Model Context Protocol server that gives your assistant read access to every channel's analytics, control of your automations, a video editor and — when you ask — live publishing and comment replies, all scoped to your own projects and authorized by a sign-in or key only you can create. You ask "how are my channels doing this month?" in plain language, and the assistant pulls the real numbers from your Klipsy account and answers.

This post explains what the Model Context Protocol is, exactly what Klipsy's server can and can't do, how access is secured, and how to connect it to your AI assistant in a few minutes.

What is the Model Context Protocol (MCP)?

The Model Context Protocol is an open standard, introduced by Anthropic in late 2024, for connecting AI assistants to external tools and data. Instead of every app inventing its own plugin format, MCP defines one: an MCP client (your AI assistant — Claude Desktop, Claude Code, or any MCP-compatible app) talks to an MCP server (a program a company runs to expose its own capabilities).

When a server is connected, the assistant can see the list of tools it offers, decide which one answers your question, call it with the right arguments, and read the result back — without you ever leaving the chat. The protocol is documented in the public MCP specification, and Klipsy implements the server half of it.

What can you do with Klipsy's MCP server?

Klipsy's server has 46 tools, and your assistant sees the ones that apply to you: 28 on most accounts, plus 17 video-editing tools once you have uploaded footage (one admin-only cost report stays hidden from everyone else). The read-only tools come first, grouped by what they help you do; the ones that change something are listed after them.

See how a channel is doing

• list_projects — list your channels/brands and their activity counts. This is the starting point: it hands the assistant the project ids the other tools need.
• get_project_performance — a 30-day trend by platform, lifetime totals, and the current best-performing template. The right first call for any "how am I doing?" question.

Dig into individual posts

• list_posts — your post inventory with per-platform publish status and current counters: views, likes, comments and shares.
• get_post_target_detail — full detail on one post on one platform: status, error message, the live URL, and when its metrics last synced. Use it to debug a single failed post.
• compare_post_metrics — put up to five posts side by side and compare their daily growth curves, to answer "why did post A beat post B?"

Decide what to make next

• list_templates_performance — your templates ranked by average views and likes per post, so you make more of what works.
• get_template_detail — one template's full configuration: prompt, voice, aspect ratio, style, duration and model.
• search_templates — find a template by name instead of needing to know its id.

Check what's running

• list_automations_status — every automation's schedule, last and next run, and any errors.
• get_pending_review_posts — drafts waiting for your approval.
• get_generation_status — poll a content generation or Studio render to see if a draft is ready.
• list_connected_accounts — connected platforms, handles, connection status and token expiry, so you can check channel health.
• get_post_comments and list_mentions — read the comments on a post (Instagram, YouTube and X) and an X account's mentions.

Tools that change something

• trigger_automation_run — run a review-mode automation now to generate its next batch of drafts. It produces content for review and does not publish or attach any social account — the drafts land in the pending-review queue.
• set_automation_status — pause or resume an automation. Pausing stops future scheduled runs immediately and never touches anything already published.
• post_to_tiktok, post_to_youtube, post_to_instagram, post_to_x, post_to_facebook and post_to_telegram — publish to a connected account, live, when you ask. They follow the same rules as publishing from the app: a content-safety check runs first, a per-account daily limit applies, and a post marked as made with AI carries the platform's own AI label where the API offers one (on by default for videos Klipsy's AI pipeline made).
• update_post — edit a live YouTube video's title, description or tags.
• reply_to_comment — post a public reply to a comment on Instagram, YouTube or X.
• The video-editing tools — analyze your footage, cut silences, trim, split and reorder clips, set captions, music and voiceover, then render the cut as a new video. Your source footage is never modified.

So you can ask things like "Which template got the best average views on my cooking channel last month, and generate a fresh draft from it," "Pause every automation on the travel project — I'm reworking the templates," or "Cut the silences from yesterday's upload and post it to TikTok and YouTube."

Guardrails: what the assistant can and can't do

This is the part worth reading twice. The publishing, editing and reply tools change real, public content, so the guardrails matter:

• It publishes only when you ask. The server tells the assistant to confirm with you before any post_to_*, update_post or reply_to_comment call, and never to repeat a publish that errored — it checks the post's status instead. trigger_automation_run only produces drafts for review; it never posts.
• Publishing keeps Klipsy's rules. The content-safety check, the daily per-account limit and the platforms' AI labels apply to the assistant's posts just as they do to yours.
• It cannot delete published posts, disconnect accounts or change your billing. Pausing an automation is fully reversible and leaves live posts alone.

The assistant reads a lot, and it acts when you tell it to, within the same limits as the app. If you want the fuller picture of how publishing and review work, how Klipsy works walks the pipeline end to end.

How is MCP access secured?

Handing an AI assistant a key to your account only makes sense if that key is tightly bounded. Klipsy's is, in four ways:

1. A sign-in or a single-purpose key. Assistants like Claude connect by signing you in to Klipsy, where you approve the connection; their access token lasts three days and refreshes for up to 30. Other clients use an MCP key, a personal access token minted with one ability — mcp:use.
2. Owner-scoped tools. Every tool is scoped to the authenticated user's own projects. There is no way to name another account's project id and read its data.
3. Plan re-checked on every request. Access isn't just granted when the key is minted — Klipsy re-verifies your plan on every single MCP call, so entitlement can't drift out of sync.
4. Revocable anytime. One connection covers all your projects. Keys are named, shown once when you create them, and revoked one at a time — Personal holds 1, Creator 3, Studio 5 and Agency 10 — and Settings → MCP → Disconnect all assistants revokes every sign-in and key at once.

Which plans include MCP?

The MCP server is available on the Personal plan and up — every paid tier. You can see what each tier includes on the pricing page. The Free plan needs no card, so you can create an account and upgrade when you want MCP.

How to connect Klipsy to your AI assistant

1. Open Settings → MCP in Klipsy and copy the server URL: https://theklipsy.com/mcp/klipsy.
2. In claude.ai or the Claude desktop app, go to Settings → Connectors → Add custom connector, paste the URL and sign in to Klipsy when asked. In Claude Code, run claude mcp add --transport http klipsy <server URL>, then /mcp to sign in.
3. For a client that takes a bearer token instead, click Create key in Settings → MCP, copy it (it's shown once) and send it as Authorization: Bearer <your-key>. Klipsy uses the streamable HTTP transport.
4. Start a chat and ask your assistant to list your projects first — that discovers the ids every other tool builds on.

From there, plain-language questions work: "How did my Shorts channel do this week?", "Which post got the most shares, and why?", "Show me drafts waiting for review."

Example things to ask

• "Give me a 30-day performance summary for each of my channels."
• "Compare my last five TikTok posts and tell me what the top one did differently."
• "Rank my templates by average views and generate a new draft from the best one."
• "Which automations have errors, and what are they?"
• "Pause every automation on my food project."
• "Reply to the top comment on my latest YouTube Short."

Why this matters

Most social tools give you a dashboard and leave the thinking to you. Pairing Klipsy's metrics with an AI assistant over MCP turns "log in, click around, export a CSV" into a conversation: the assistant reads views, likes, comments and shares straight from your account, spots the pattern, and acts when you ask without you touching the UI. It's the same create → publish → measure loop Klipsy already runs, now queryable in natural language. For how that loop fits together, see why Klipsy runs the whole loop and the full feature tour.

FAQ

Which AI assistants work with Klipsy's MCP server?

Any MCP-compatible client — claude.ai and the Claude apps, Claude Code, the ChatGPT desktop app, and other assistants that support the Model Context Protocol. Klipsy runs a standard streamable-HTTP MCP server: Claude signs in, and clients that take a bearer token connect with an MCP key.

Can the AI post to my social accounts?

Yes, when you ask it to. The post_to_* tools publish live to TikTok, YouTube, Instagram, X, Facebook Pages and Telegram accounts you connected, with the same content-safety check, daily limit and AI labels as the app. The assistant is told to confirm with you before posting, and it never posts on a schedule of its own — automations still run on the cadence and review mode you set.

Is my data safe if I connect an AI assistant?

Every tool is scoped to your own projects, your plan is re-checked on every request, a key carries a single ability (mcp:use), and you can revoke any sign-in or key at any time from Settings → MCP.

What metrics can the assistant read?

Views, likes, comments and shares — per post, per platform, and rolled up per project — the same numbers Klipsy tracks natively.

What does MCP access cost?

It's included on the Personal plan and every tier above it. See the pricing page for details.