Claude Code for Social Media: Plan, Write, Design and Post From Your Repo

Claude Code is Anthropic's agent for software work. Anthropic's overview describes it as a tool that reads your codebase, edits files, runs commands and connects to your development tools, and it runs in the terminal, in VS Code and JetBrains, in a desktop app and in the browser. None of that sounds like social media. The link is the folder it works in.

If you make software, the best raw material for your social media is already sitting in your repository: the changelog, the release notes, the commit messages between two tags, the screenshots folder, the roadmap file, the issues users opened. A social media manager at a bigger company spends hours asking engineers what shipped. Claude Code can read the answer directly, and it can tell a refactor nobody will notice from a new widget people asked for.

So "Claude Code for social media" is mostly three jobs. First, finding the story: reading what changed and picking what a user would care about. Second, writing and designing: turning that into posts in your voice, one version per network, with real pictures. Third, getting it onto a calendar: saving drafts or scheduled posts in a tool that is logged in to Instagram, LinkedIn and the rest. Claude Code does the first job on its own. The second and third need a connection to a social media tool, which is what most of this page covers.

This page is about the terminal and IDE agent. If you use Claude in the chat app on the web, desktop or phone, read Claude for social media marketing instead, which covers Projects, connectors, Claude Design and Cowork. If you only want the click by click path to a first scheduled post, the guide on how to schedule social media posts with Claude walks through it. Here we go deeper on the parts only Claude Code has: CLAUDE.md, skills, slash commands, hooks, permission rules and headless runs.

Not by itself. A fresh Claude Code install has no login to any social network. It can write a perfect LinkedIn post into a file, but it cannot press publish, hold a post until Tuesday at nine or read last week's comments. Searches like "can claude code post on social media" usually come from people who expected it to, so it is worth being plain about where the posting comes from.

It comes from MCP, the Model Context Protocol. An MCP server is a service that gives Claude a set of tools, such as "save a draft", "schedule a post" or "read analytics". You add the server to Claude Code once, and from then on Claude can call those tools when you ask. Claude Code's MCP documentation recommends the HTTP transport for remote servers and shows how to add one with a single terminal command.

A social media MCP server holds your network connections on its side. You connect Instagram or LinkedIn inside that tool, through each network's own sign in, and Claude never sees those passwords. When you ask for a post, Claude calls the tool, and the tool saves, schedules or publishes it.

Spreadcast is one such MCP server, built for people who make App Store apps. Connected to Claude Code, it gives Claude tools to design posts in your app's brand, write a caption for each network, check posts against each network's rules, save drafts, schedule and publish (on a paid plan, once you allow it) and read results.

The first is about shared files. A project scoped server sits in .mcp.json, which you commit. The MCP documentation supports environment variable expansion in that file, so a header can read the key from a variable on each person's machine rather than from the file itself. It also notes that in headless runs, Agent SDK sessions and cloud sessions Claude Code cannot show its approval prompt and loads project scoped servers without asking.

The second is about output size. The same page says Claude Code warns when one MCP tool returns more than 10,000 tokens and caps a result at 25,000 tokens by default. That is why Spreadcast pages its post lists and shows compact previews in the session while the full 1080x1350 pictures stay in your account.

Every Claude Code session starts with an empty memory, and CLAUDE.md is how you give it a lasting one. Anthropic's memory documentation describes CLAUDE.md as instructions you write that load at the start of every session. A project file sits at the root of the repository, or in the .claude folder, and is shared with the team through version control. A personal file in your home folder applies to all your projects, and a CLAUDE.local.md in the project root holds private notes you keep out of git.

For social media that means your brand voice can live next to your code and travel with it. A new contractor clones the repository and Claude Code already knows that Streakly never uses exclamation marks, calls its users "people building a habit" rather than "users", and posts to Instagram, LinkedIn and Threads.

Keep it short. The same documentation recommends staying under 200 lines per CLAUDE.md file, because longer files use more context and Claude follows them less reliably. You can pull in a longer brand guide with an import, written as an at sign followed by the path, and imports can nest up to four levels deep. Imported files still load at launch, so they cost the same context as text in the file itself.

Anthropic is also clear about what CLAUDE.md is not. The memory page says Claude treats these files as context, not enforced configuration, and points to a hook when something must be blocked regardless of what Claude decides. For social media that line is the important one: "never publish without asking" belongs in CLAUDE.md as a reminder and in a hook or permission rule as the actual guard.

If you have no CLAUDE.md yet, run /init in the project. Claude reads the codebase and writes a starting file with build commands and conventions, and if one already exists, it suggests changes instead of overwriting. Then add a social media section by hand, because Claude cannot discover your voice from the code.

"Claude code social media manager skills" is the most common search in this whole topic, and it mixes two meanings. Some people want to know which skills a social media manager needs to use Claude Code. Most want ready made skills, the Claude Code feature, that make it act like a social media manager. This section covers the second, because it is where the time savings are.

A skill is a folder with a file named SKILL.md inside. Claude Code's skills documentation lists the places one can live: a personal folder under .claude in your home directory for all your projects, or the project's own .claude/skills folder for one repository and everyone who clones it. The file opens with a short header, and the description in that header is what Claude reads to decide when the skill fits.

Custom slash commands used to be a separate feature. The same page now says commands have been merged into skills: a file in .claude/commands and a skill folder with the same name both create the same slash command, and old command files keep working. So a "slash command for LinkedIn posts" and a "LinkedIn skill" are the same thing today. The skill form adds a folder for supporting files, such as example posts.

Good social media skills are narrow, one per job you repeat. Three are worth writing first for an app maker. A release skill that reads the changes between two tags and drafts a carousel and captions. An Instagram carousel skill that knows your slide pattern: a cover line under eight words, one slide per change, a last slide that says where to get the app. And a LinkedIn post skill that writes in first person from the founder, explains why a change matters and ends with a question rather than a link.

Two skills come up in searches again and again: an Instagram carousel skill and a LinkedIn post skill. Both are worth writing, and both have the same weak spot. A skill can describe a carousel perfectly, but it cannot draw one. Without a design tool, a "carousel skill" produces slide text and maybe an HTML mock that you still have to turn into images by hand.

For Instagram, write the slide pattern into the skill and let the connected tool do the drawing. Instagram's own rules shape the pattern. Meta's content publishing guide allows up to 10 images, videos or a mix in one carousel, accepts JPEG as the only image format and crops every image in a carousel to the first one, with a square crop by default. A skill that calls Spreadcast's carousel tool gets 3 to 8 slides at 1080x1350 in your app's colors, fonts and logo, with real screenshots in an iPhone frame, saved as JPEG pictures ready for Instagram.

For LinkedIn, the skill matters more than the picture. Tell it who posts (the founder, in first person), how long the post runs, how the first line should read before LinkedIn cuts it off, and to end without a link in the body. Then have it ask the tool for LinkedIn's posting rules before writing, so length and format checks happen before the draft is saved.

A note on skills from GitHub. Many community repositories ship Claude Code social media skills, and some are good starting points. Read every file before you copy one in: a skill can tell Claude to run scripts or install software, and it runs with the permissions of your session. Copy the pattern, not the whole folder.

Skills are not the only way to get a repeatable command. An MCP server can publish its own prompts, and Claude Code's MCP documentation says they appear as commands: type a slash and they show up in the list as the server name, a colon and the prompt name, marked MCP. Arguments go after the command, separated by spaces.

Spreadcast ships two. "Draft Posts From My App's News" checks your account, lists new moments from your App Store listing, such as a new version or a rating milestone, designs a picture for each, writes captions for your connected networks and saves each as a draft for approval. "Plan A Week Of Posts" reads your brand kit, moments and the last 30 days of results, then drafts five posts by default (or the number you give it), spread over the week at the times that worked best.

In Claude Code they show up as /mcp__spreadcast__spreadcast_post_my_moments and /mcp__spreadcast__spreadcast_plan_week if you named the server spreadcast. They are a quick start. Once you know what you want, copy the prompt into your own skill and add your voice rules and the files Claude should read from the repository.

Instructions shape what Claude tries to do. Hooks and permission rules decide what Claude Code allows. Anthropic's permissions page puts it directly: permission rules are enforced by Claude Code, not by the model, and instructions in your prompt or CLAUDE.md do not change what is allowed. For a tool that can put words on your public profile, that difference is the whole point.

Permission rules come first because they need no code. Each MCP tool has a name made of mcp, the server name and the tool name, joined by double underscores. The permissions page explains that rules are checked in the order deny, then ask, then allow, and that the first match wins. So you can allow every Spreadcast tool whose name starts with get or list to run freely, put the tools that create, approve, reschedule or delete posts under ask, and Claude Code will stop and show you each one before it runs. You manage these rules with /permissions or in settings.json.

Hooks go further. Anthropic's hooks guide describes them as shell commands that Claude Code runs at fixed points in its lifecycle, which gives you deterministic control rather than relying on the model to choose. A PreToolUse hook runs before a tool call and can block it: a script that exits with code 2 stops the call and passes its message back to Claude. The guide's matcher examples show MCP tools matched by the same double underscore names, so one hook can watch every Spreadcast tool.

Useful hooks for social media are small. A hook that blocks the approve tool outside your working hours. A hook that writes every saved draft to a log file in the repository, so you have a record of what Claude prepared. A hook that refuses any post whose caption contains a word from your banned list. A Notification hook that pings your desktop when Claude is waiting for you to approve something.

For most social media work, routines are the right scheduler. Anthropic's routines page describes a routine as a saved prompt with one or more repositories and connectors that runs on Anthropic's cloud, so it keeps working when your laptop is closed. It can start on a schedule (hourly, daily, weekdays or weekly), from an HTTP call, or from a GitHub event, and GitHub release events are on the supported list. Routines are on the Pro, Max, Team and Enterprise plans and are still a research preview, so details can change.

Two facts on that page shape how you should set one up. Routines run autonomously, with no permission prompts, and Claude can use every tool from an included connector, including writes, without asking. And the connectors a routine sees are the ones on your claude.ai account: servers you added in the terminal with claude mcp add live on your machine and do not appear. So for a routine, add Spreadcast as a custom connector in Claude on the web, choose Make Drafts on the consent screen, and leave publishing off. A routine that can only save drafts cannot surprise you.

A good weekly routine prompt is short and self contained: "Read the commits merged since last Monday and the open roadmap issues. Using Spreadcast, read my brand kit and last week's results, then save four drafts for this week in my voice, one per network as usual. List them at the end." Set it a few minutes past the hour, such as 9:07, since the page notes a run set exactly on the hour can start late.

Claude Code can also run without the interactive screen. Anthropic's page on running Claude Code programmatically shows the print flag, written claude -p followed by your prompt, which runs one task and exits with a success or failure code your script can check. You can approve tools in advance with --allowedTools. The same page recommends a bare mode for scripts, which skips CLAUDE.md, skills, hooks and MCP servers unless you pass them in and needs an API key rather than your subscription login.

GitHub Actions wraps the same engine in a workflow. The Claude Code GitHub Actions page says that when a workflow gives the action a prompt, Claude runs without waiting for a mention, on any GitHub event, including a cron schedule. The prompt can be a skill from the repository's .claude/skills folder, so a workflow that runs /social on every published release reuses the skill you already wrote. Authentication is either an API key or a long lived token from your Claude subscription, which the page says works on Pro, Max, Team and Enterprise plans.

The GitHub page lists the costs plainly: each run uses GitHub Actions minutes and Claude tokens, or your subscription if you use the token. It also notes that GitHub runs scheduled workflows only from the default branch and, in public repositories, switches the schedule off after 60 days without activity. For a social media key in a workflow, store it as a repository secret and use a Spreadcast key that cannot publish.

Yes, through a tool that is connected to Instagram, not directly. You connect Instagram inside a social media tool once, connect that tool's MCP server to Claude Code, and Claude asks the tool to save or publish.

Instagram sets the rules every tool works within. Meta's content publishing guide covers publishing for Instagram professional accounts, so a personal account has to switch to a professional one first. It limits each account to 100 posts published through the API in a moving 24 hour window, with a carousel counting as one post. And it fetches each picture from a public address at the moment of publishing, which is why a tool hosts your images for you rather than reading them from your laptop.

Can Claude Code post to Instagram through Spreadcast? Claude can design the post or carousel, write the caption with hashtags, check it and save it as a draft. With Spreadcast Pro or Studio and a connection you allowed to publish, it can also schedule or publish, and Spreadcast tells Claude to show you the post, network and time and wait for your yes first. Reading results covers likes, comments and follower counts; Instagram reach is not among them.

What it does not do: Spreadcast does not read or send Instagram DMs, and it does not scrape other accounts. Searches for a "claude code instagram scraper" lead to projects that log in as you through a browser, which is a different kind of tool with different risks. The Instagram MCP server page compares the kinds of Instagram servers, what Meta's API allows and what to check before connecting one.

Yes, the same way: through a connected tool. The open LinkedIn permission for posting is w_member_social, which LinkedIn's Share on LinkedIn guide describes as required to create a post on behalf of the member who signed in. The same guide lists a daily limit of 150 requests per member, counted in UTC.

Spreadcast uses that permission, so posts Claude Code drafts for LinkedIn go out from your personal profile once you approve them. Company pages are not supported today, because posting as a page needs a separate permission LinkedIn grants through its own review. For the full picture of LinkedIn servers, including the cookie based ones on GitHub and why they put accounts at risk, see the LinkedIn MCP server page.

Where Claude Code earns its place on LinkedIn is the writing. A founder post works best when it explains a decision, and Claude Code can read the pull request discussion and the issue behind a change, which gives it the story rather than just the feature name.

"Claude code social media agent" and "social media dashboard claude code" are real searches, and Claude Code is good at building both. It can write a small app that calls the Instagram and LinkedIn APIs, stores tokens, schedules posts and charts results. The question is whether you want to own that app.

Here is what you take on. A Meta developer app with Instagram permissions, reviewed by Meta, publishing only to professional accounts, with images hosted at a public address and the 100 posts per 24 hour limit from Meta's guide. A LinkedIn developer app with the Share on LinkedIn product added, and the per member request limit from LinkedIn's guide. Token storage and refresh for every network. A queue that publishes at the right time when your laptop is off. A way to design the pictures. And keeping all of it working each time a network changes its API.

That is reasonable if publishing is your product. For an app maker who wants posts out every week, connecting an existing server is faster, and Claude Code spends its time on the story rather than on OAuth flows. If you are comparing servers, the best social media MCP servers ranks the options with prices, and the AI agent for social media page looks at agents more widely.

Social media work is light next to coding. A coding session carries file contents, test output and many steps; drafting four posts reads a changelog, calls a handful of tools and writes a few hundred words. Anthropic's cost page gives a sense of the coding side: across enterprise deployments, the average is around $13 per developer per active day and $150 to $250 per developer per month. A weekly social media routine sits far below that, though nobody can give you an exact number for your setup.

The same page lists habits that keep usage down: keep CLAUDE.md short, move long procedures into skills that load only when used, start a fresh session between unrelated jobs, and turn off MCP servers you are not using in /mcp.

Run /usage to see where it goes: on paid plans the cost page says it breaks usage down by skills and individual MCP servers and lists the heaviest scheduled tasks. For most solo founders Pro is enough; Max makes sense if you also code all day. Your Claude plan is separate from the social media tool's plan, so budget for both.

For the connection itself, the Spreadcast MCP server reference lists every tool Claude gets and how sign in works. For the chat app side of Claude, read Claude for social media marketing, and for the shortest route to a first scheduled post, the guide on how to schedule social media posts with Claude. If you also use ChatGPT, ChatGPT social media automation covers that side.

And if the hard part is deciding what to post at all, how to promote an app on social media covers the content that Claude Code will then help you produce.