Post, schedule and run ads from Claude Code

Claude Code is Anthropic’s terminal agent. It lives in a working directory, reads and writes files, runs commands, and installs plugins from a marketplace with one command. That makes it the natural home for the parts of marketing that are really pipelines: render, check, publish, log.

Hermoso is the marketing half: 830+ tools over one OAuth’d MCP endpoint covering competitor ad research, finished image and video ad generation across 90+ models, publishing and scheduling to ten social channels, and campaign management on Meta, Google Ads, Microsoft Advertising, Reddit Ads, LinkedIn, Pinterest, X Ads, TikTok Ads, Snapchat Ads, ChatGPT Ads and Apple Search Ads. One connection, and Claude Code has all of it.

Connect Hermoso to Claude Code

1. One command, and no key.

   claude plugin marketplace add hermoso-ai/hermoso && claude plugin install hermoso@hermoso

   Already inside a session? Type /plugin marketplace add hermoso-ai/hermoso, then /plugin install hermoso@hermoso. It installs four Hermoso skills (research, ad-from-brand, generate, product-photoshoot) that run the hermoso command through npx, so no tool list is loaded into your sessions and nothing costs context until it runs.
2. Sign in once.

   npx -y hermoso auth login

   It opens a browser sign-in on loopback, the way gh does, and stores an owner-only credential in ~/.hermoso/config.json. Claude Code runs it for you the first time it needs Hermoso. No browser on that machine, say CI, a remote box or a container? Create a key under Settings → Agents & API keys and run npx -y hermoso auth login --token hmk_…. A key carries real spend authority, so treat it like a password.
3. Optional: the MCP server as well. If you also want Hermoso’s tools in Claude Code’s own tool list, run claude mcp add hermoso -- npx -y hermoso mcp after signing in. It reads the same stored credential, so no key goes into .claude.json. The skills reach every tool without it.
4. Ask Claude Code for something. “Pull my three closest competitors’ longest-running ads and remix the best hook into a 9:16 video,” or “post this render to the channel.” The agent picks the tools; you approve the spend.

The plugin and the optional MCP server reach the same tools and the same account. The difference is context: a tool list is loaded into every session before you type anything, while the skills cost nothing until a command runs.

Every channel Claude Code can publish to

Each of these has its own page with that channel’s real caption cap, its real visibility options, what it refuses and why, and a worked example phrased for Claude Code.

Every one of them queues through one schedule_post calendar with per-channel captions, and list_scheduled reports per-channel outcomes rather than a single green tick — “the job succeeded” and “the post went out” are different questions.

Two channels that are built but not offered

Being straight about this is worth more than a longer list. Two publishing surfaces exist in the codebase and are held back, each waiting on the platform rather than on us:

• Reddit: organic posting and post stats are built, but Reddit’s Responsible Builder Policy requires explicit written approval before an app may use the Data API, and that ticket has not been granted. The connector is held back rather than offered as an OAuth screen leading to a dead API. Reddit Ads is a different connection and it is live: the full campaign tree, forecasting, targeting search, pixels and lead forms.
• Google Business Profile — post_to_google_business is built, with events, offers and a real call-to-action choice, and so are the listing insights. Google allowlists the Business Profile API per project and the default quota is zero, so every call would 403 for every user. It stays gated until Google grants the project access.

A connector that would hand you a successful OAuth and a dead API is worse than one that is not offered yet, so neither is listed as available.

Three things to type once it is connected

1. Research. “Who are my three closest competitors, and what are their longest-running Meta ads? Tell me which hook keeps repeating.” — walks find_competitors → pull_competitor_ads → mine_angles.
2. Create. “Make me a 9:16 UGC video ad using that hook, with my real product photo and captions burned in.” — walks get_brand → plan_ad → render_ad, and hands back a finished mp4 URL.
3. Ship. “Schedule it to Instagram and TikTok for Tuesday 9am, and build a paused Meta campaign around it targeting the US at $30 a day.” — walks schedule_post → create_meta_campaign → create_meta_ad.

What Claude Code cannot do on its own

An agent with a budget is only useful if the blast radius is bounded, so the fence is in the server rather than in a prompt. Every ad campaign, ad set, ad group and ad is created paused, and the status switches that arm real money refuse to run without an explicit confirmation flag. Every render is quoted first — hermoso_capabilities publishes each model’s exact credit cost before anything spends — and credits are reserved before the first provider call and settled at the exact cost afterwards, so a failed dispatch refunds rather than leaving you billed for nothing. Deletes are confirm-gated too, and every campaign tree is read back from the ad platform before your agent tells you what it built.

One step is deliberately not headless: linking an account is an OAuth consent screen, so you connect the channel once in a browser. Everything after that is a tool call.

Beyond publishing

Publishing is the last step, and it is the smallest one. The same connection gives Claude Code the Meta, Google and LinkedIn ad libraries plus organic TikTok, Instagram, YouTube, Reddit and Threads for research; 50+ image, video, voice and writing models on a single credit meter, with UGC avatars, dubbing, upscaling and native HTML ad formats; and the full campaign tree on eleven paid platforms. There are no per-vendor API keys to hold and no separate app to learn.

See the full tool list, or what a marketing MCP is. For the longer story on this client specifically, see Claude Code marketing.

Other clients: Claude · ChatGPT · Cursor · Codex · Cline · OpenClaw · Grok Bot · Hermes · any MCP client

Frequently asked questions

Can Claude Code post to social media?

Yes, once you install Hermoso. Claude Code can then publish to Facebook, Instagram, Threads, TikTok, YouTube, X, LinkedIn, Pinterest, Bluesky and Telegram — and schedule to all ten from one queue with per-channel captions.

How do I add Hermoso to Claude Code?

One command: claude plugin marketplace add hermoso-ai/hermoso && claude plugin install hermoso@hermoso, or inside a session /plugin marketplace add hermoso-ai/hermoso then /plugin install hermoso@hermoso. It installs four Hermoso skills that drive the hermoso CLI, so no tool list is loaded into your sessions. Sign in once with npx -y hermoso auth login, which opens a browser and stores the credential itself, so there is no key to create and none to paste. The plugin and the optional MCP server reach the same tools and the same account. The difference is context: a tool list is loaded into every session before you type anything, while the skills cost nothing until a command runs.

Can Claude Code generate the video and images too?

Yes, natively rather than by brokering it out. 50+ image, video, voice and writing models sit on one credit meter, and Claude Code can price a render before it runs it because hermoso_capabilities publishes each model's exact cost. Output is a finished ad — your real product in the scene, captions and CTA composited over the render.

Is there a free tier for Claude Code?

Yes. A free account starts with 30 credits and 250+ more you can earn as you set up and create, and the MCP server itself is free to add to Claude Code. Paid plans are $19, $49 and $149 a month for 1,000, 3,000 and 10,000 credits. Every feature is on every plan including Free, and the only exception is enterprise SSO.