Metricool MCP: The Claude and ChatGPT Connectors, Tools, Limits and Setup

The Metricool MCP is a remote server, run by Metricool at the address ai.metricool.com/mcp, that lets an AI assistant work inside your Metricool account. Once it is connected you can ask Claude or ChatGPT, in ordinary sentences, which Reel did best last month, how your engagement compares with three competitors, or to put a post on Instagram and LinkedIn at the best hour for each. Metricool's MCP page presents it as a way to read analytics, plan posts and check rivals from one chat instead of opening the dashboard.

It has two front doors. On September 10, 2026, Metricool published a press release, Metricool launches official connectors for Claude and ChatGPT, saying it is now listed as an official connector in both assistants and that every Metricool user, free accounts included, can use them. Behind both connectors sits the same MCP server, which Claude Code, Cursor, Gemini CLI, n8n, Make, Zapier and other clients reach directly by its address.

MCP is short for Model Context Protocol. The official MCP introduction describes it as an open standard for linking AI applications to outside systems, so one server works in many assistants instead of needing a separate plugin for each. In practice the assistant does the thinking and the talking, and Metricool supplies the data and does the scheduling. The server can only reach what your Metricool account already holds: the brands, networks, numbers and calendar you set up there.

Its strength is reading. Most of its tools pull numbers and past posts out of Metricool, and that is where it beats most other social media servers. It adds no design studio and knows nothing about your product beyond what your accounts have posted, which the sections below come back to. For Metricool as a whole product, see our Metricool comparison for app founders, and for a ranked field of other tools, the best Metricool alternatives.

Metricool's limits article is plain about one thing: connecting an assistant does not lift any plan limit. An action through Claude or ChatGPT counts exactly like the same action in the Metricool app. The article adds that when an action fails or the data looks thin, the plan limit is the first thing to check, before you suspect the connection.

Here is a worked example with round numbers. Alex runs a small brand on Metricool Free with Instagram, TikTok and Threads connected. On the first of the month Alex asks Claude to plan a week: one post a day on each of the three networks. That is 21 posts, and the free plan allows 20 a month, so the last one fails, and there is nothing left for the rest of the month. A single multipost to three networks may count as one post or as three; Metricool's pages do not say, so plan as if each network counts.

The analytics window matters as much. Free keeps 30 days of numbers. A question such as which month this year had the best engagement has no answer on Free, because the older months are not stored. The assistant will either say so or answer from the 30 days it can see, and it is worth asking which one it did.

LinkedIn is the other trap. A free account cannot connect LinkedIn at all, so an assistant asked to post there will find no LinkedIn profile on the brand. For a LinkedIn workflow through the Metricool MCP the realistic floor is Starter.

Competitor analysis is the part of the Metricool MCP that few other social media servers have. Metricool lets you add rival profiles to a brand, and two tools expose them: one lists the competitors you added, the other fetches and analyzes their posts. Metricool's MCP basics article describes what that allows: comparing growth, engagement and posting frequency, and asking for strategy suggestions from the result.

The size of the field depends on the plan. The pricing page gives Free 5 competitor profiles and Starter up to 100. Metricool's Python package description names the networks the competitor tools cover: Instagram, Facebook, X, Bluesky, YouTube and Twitch. TikTok, LinkedIn, Pinterest and Threads are not on that list, so a question about a rival's TikTok will not get Metricool data.

A useful request is narrow and dated. Metricool's guide to social media management with Claude, published on September 4, 2026, gives the shape: compare my engagement rate with my three main competitors on Instagram over the last 30 days. Asking for a period inside your plan's history, on a network the tools cover, avoids most empty answers.

Treat the conclusions as a starting point. The assistant sees posts and public numbers, not why a rival's post worked, and a sample of a few weeks can mislead. The data is the valuable part; check the story the assistant tells about it.

Posts made through the MCP go into Metricool's Planner, not straight to the network. Metricool's Claude guide explains that Claude hands the post to Metricool, which then publishes it on time or, for formats it cannot publish on its own such as some Instagram Stories, prompts you to finish it by hand. Metricool's MCP page lists three timing choices: publish now, a date and time you pick, or the best time from your own data.

There is no separate draft mode in the tool list. A scheduled post is live on the calendar and will publish at its time unless you change it. Metricool's scheduling article suggests opening the direct link in each answer to check the post, its attachments and its formatting in the Planner before it goes out, and notes that some formatting or attachment changes may still need the Planner itself.

Review is the safer path. Metricool's limits article lists creating posts for review and sending them for review among the MCP's Planner actions. A post sent for review waits for approval and, per Metricool's approval system article, a rejection always stops it from publishing, and editing it resets the review. That same article puts the approval system on the Advanced and Custom plans, and the MCP articles do not say whether review requests work below Advanced, so expect the plan rule to apply.

Carousels work when the images exist. Metricool's own Claude video schedules a carousel to Instagram and LinkedIn with the images already attached, and its Claude guide says videos can come from a link or an uploaded file. No tool in the list makes an image. The pictures come from you, a design tool, or whatever your assistant can produce on its own, and Metricool's limits article says Canva and Google Drive files are out of the MCP's reach, even though Metricool's own planner connects to both on Starter and above.

People searching for the Metricool AI assistant and the Metricool MCP often mean different things, and the two sit on opposite sides. The Metricool AI social media post generator is built into Metricool's planner. You pick a topic, a tone and a network, and it writes the post, adapts the text for each network, writes calls to action, changes the length, translates and proofreads, and suggests hashtags for TikTok and Instagram. The page describes text only, with no image generation.

It runs on credits. The pricing page gives 5 credits per brand a month on Free, 20 on Starter and 35 on Advanced, with custom amounts on Custom. So is Metricool an AI tool? Partly: it is a scheduling and analytics tool with an AI writer inside it.

The MCP is the reverse direction. Your own assistant, Claude or ChatGPT, does the writing and the reasoning in its own window, with the model you already pay for, and uses the MCP to read your Metricool data and put the result on your calendar. It does not spend Metricool's AI credits, and it can write as much as your assistant plan allows.

Neither makes the picture. If writing is the slow part, either helps. If the slow part is the image, or knowing what to post about in the first place, something else has to sit in the chain.

Metricool keeps the two apart. Metricool's article on MCP and API access describes the MCP as a connector for plain language requests in AI clients, open on every plan, and the API as a programmatic interface for HTTP requests from tools such as n8n, Make, Zapier or Postman, open only on Advanced and Custom. API calls carry an access token in a header, plus your user ID and brand ID on every call, and the API can export data the MCP cannot reach, such as inbox comments.

That split explains a pattern in the setup guides. Sign in by OAuth and you are on the MCP, which works on Free. Paste a token or an API key, as the n8n route, the Claude Desktop key variant and TypingMind's token option do, and you are on the API, which needs Advanced. If a guide you found online starts by asking for your user token and user ID, it predates the hosted server or targets the API route.

On the code side, Metricool's MCP FAQ says the server is open source in a public GitHub repository and published as a Python package. The mcp metricool package on PyPI lists Metricool staff as its maintainers and describes a local setup that runs through uv with your Metricool token and user ID, and a requirement of API access on the Advanced tier. Most people no longer need it: the hosted server at ai.metricool.com/mcp does the same work after a browser sign in. Several unofficial Metricool MCP projects also exist on GitHub; a server that asks for your Metricool password rather than a Metricool sign in or a token is not Metricool's.

Metricool's troubleshooting article covers seven cases. The first check is always the address: ai.metricool.com/mcp exactly, starting with https, with no extra path, space or different host. The second is the method, since each client has its own: OAuth by address, a header token, a terminal command, or the desktop config file.

When a client never finishes connecting, the sign in usually stopped halfway. Finish the browser step and click Grant access. In Cursor, click Connect after installing, because the server shows as unauthenticated until you do. In Gemini CLI, run /mcp auth metricool. In Claude Code, run /mcp, choose Metricool and authenticate. If Claude Code rejects the add command, check that it is installed and that your Claude plan is Pro or Max.

If Claude Free refuses to add Metricool, another connector already uses the single free slot; remove it and try again. If Cursor shows no install flow, you are probably in a browser rather than the desktop app. If n8n fails, check that the credential uses Header Auth with your API key from Account Settings, API.

The last case is the one people misread most: the connection works but an action fails or the data looks short. Metricool says that is usually a plan limit, not a broken connection, such as the 20 post monthly cap or the 30 day analytics window on Free.

Metricool's MCP page says the MCP uses the same permissions and sign in system as your Metricool account, and that your data is not stored in the AI model. That is the vendor's claim about its own side. What the assistant does with the data in your chat is governed by your assistant's own settings and plan, so read those too.

Revoking needs care, because Metricool's pages differ. The MCP page says you can revoke access at any time from your settings, while the troubleshooting article says to remove the connector in the AI client where you authorized it, because access cannot be revoked from inside Metricool. The safe habit is to remove it in the client: in Claude, open Settings, Connectors, the menu next to Metricool and Disconnect.

The permissions screen deserves a careful read for a different reason. Metricool's guides say it describes broader access than the MCP actually uses. The tool list, not the consent screen, is the real map of what an assistant can do.

Keep your client's approval prompt on for scheduling and editing. Every write goes onto a live calendar, a publish now request goes out at once, and on plans without the approval system there is no review step between the assistant and your audience.

Spreadcast is a social media tool for people who make apps on the App Store, and it runs its own MCP server at spreadcast.app/api/mcp. Where Metricool's server reads numbers and puts posts you bring onto a calendar, Spreadcast's server makes the post. The assistant reads your app's brand kit and designs the picture in your colors, fonts and logo, with real screenshots of your app in an iPhone frame, in 11 kinds of post, or a carousel of 3 to 8 slides at 1080 by 1350. It makes no AI photos and no video. It writes a caption for each network, checks the post against each network's rules and saves it as a draft.

It also knows what is worth posting. Spreadcast reads your App Store listing and finds moments: a new version with what changed, taken from your release notes, or a rating milestone worth a thank you. Two ready prompts start from there: Draft posts from my app's news, and Plan a week of posts. You are not inventing topics in the chat.

Drafts are the default by design. When you connect Claude or ChatGPT you sign in with Spreadcast, and the consent screen asks one question with two answers: Make Drafts or Make and Publish. With Make Drafts, the assistant can design, write and save but cannot schedule or publish, however the request is worded. Approving a draft is its own step. Claude Code and Cursor can use a key from Settings, AI Assistants instead, which is draft only unless you turn on Can Publish Posts.

Spreadcast publishes to Instagram, Facebook Pages, LinkedIn personal profiles, Threads, Pinterest, TikTok, YouTube, Bluesky, Telegram channels, Discord channels and Mastodon, and reads back likes, comments and follower counts. On TikTok, posts stay private until TikTok audits the app. The Free plan includes the MCP server and saves drafts. Pro, at $29 a month, adds scheduling and publishing, with the design and the captions included in that price. Studio, at $79 a month, covers 5 apps, each with its own brand, and 3 seats. The full tool list is on our social media MCP server page, and the MCP overview gives the short version.

Metricool and Spreadcast are two of about ten tools with a social media MCP server today. Buffer, Postiz and Hootsuite are among the others. Our roundup of the best social media MCP servers puts them side by side, with free plans, sign in methods and what each can and cannot do, and the Buffer MCP server guide goes as deep on Buffer as this page does on Metricool.

For the assistants themselves, see Claude for social media and ChatGPT social media automation, plus the guide to scheduling social media posts with ChatGPT.