Give your AI agent a marketing department

An AI agent can only do what it has tools for. Give Claude, Codex or Cursor a marketing brief today and it will write you a plan; it cannot look at what your competitors are actually running, it cannot render a video ad with your product in it, and it certainly cannot put that ad live on your Instagram. Hermoso is the connector that changes what the answer to "make me some ads" can be. It turns the assistant you already use into an AI marketing agent.

Add one OAuth'd MCP server (https://app.hermoso.ai/mcp) and your agent gains a marketing department's worth of callable tools: ad research, creative generation, post-production, publishing, scheduling, and campaign management across twelve paid platforms. No API keys to juggle, no per-vendor accounts, one credit meter.

What your agent can actually do

• Research. Find a brand's competitors, pull their live ads out of the Meta, Google and LinkedIn ad libraries, rank them by how long each has kept running, tear down the winners beat by beat, and mine the angles that repeat. Plus organic TikTok, Instagram, YouTube, Reddit and Threads.
• Create. Draft a brand from a website, a description or a social handle. Plan an ad, then render it: finished video and image creative across 100+ models, with your real product, copy, casting and call to action baked in. UGC avatars, voiceover, native HTML ad formats where the type has to be pixel-exact.
• Fix. Remix a static into motion, recast the actor, reframe to another aspect ratio, upscale, dub into another language, swap the voice, repair a single beat, stitch acts together, clip a long video into short-form cuts.
• Publish and schedule. Facebook, Instagram, Threads, TikTok, YouTube, X, LinkedIn, Pinterest, Bluesky, Telegram and Google Business Profile, on one calendar, up to a year ahead, or dropped into the brand's posting queue to take its next free slot.
• Run the media. Build and read campaigns on Meta, Google Ads, Microsoft Advertising, Reddit Ads, LinkedIn, Pinterest, X Ads, TikTok Ads, Snapchat Ads, ChatGPT Ads, Apple Search Ads and AppLovin Ads: campaigns, ad sets, ad groups, ads, keywords, budgets, targeting, audiences and performance.
• Remember. Brand memory, reusable skills, per-client profiles and team seats, all readable and writable by the agent so it does not start from zero every session.

Built so an agent cannot hurt you

Autonomy is only useful if it is safe, and the safety here is structural rather than advisory.

• Every campaign on every paid platform is created paused. The only tools that can arm real ad spend are the status switches, and each of them refuses to run without an explicit confirmation flag. Creating an enabled child under an already-enabled Google Ads parent is held to the same gate, because that serves on the very next auction.
• Every campaign tree is read back from the ad platform before the agent describes it, so what you are told is what the platform did, not what was requested.
• Deletes are confirm-gated. Generation reserves an estimate and settles the exact cost, and a failed dispatch refunds.
• No password reaches the model. Every channel is connected through its own OAuth flow in your Connectors settings.
• You tick exactly which Pages, ad accounts and boards Hermoso may use. The server re-verifies every id against the live connection, and an unticked account is unusable, including by the agent and including for its own owner.

Every AI agent Hermoso works in

Each agent below has its own setup guide with the exact steps, what that agent asks before it acts, troubleshooting, and every social channel it can post to. Every route was read from that vendor’s own documentation, and each guide says which steps we ran ourselves.

Three ways an agent connects

• A connector URL. Chat apps and hosted agents add https://app.hermoso.ai/mcp and sign in with OAuth. Hermoso supports dynamic client registration, client ID metadata documents and PKCE, so there is no key to create and nothing to paste.
• A plugin or extension that drives the CLI. Claude Code, Codex and Gemini CLI install Hermoso’s five skills in one command. The skills run the hermoso command, so no tool list sits in your agent’s context until a command runs.
• A local server. Any agent that can start a process runs npx -y hermoso mcp after one hermoso auth login, which stores the credential so no key goes in a config file. An agent that can do none of these sends an API key (hmk_…) in an Authorization header.

Install in one line

In a coding agent it is one command. Claude Code: /plugin marketplace add hermoso-ai/hermoso, then /plugin install hermoso@hermoso. Codex: codex plugin marketplace add hermoso-ai/hermoso, then codex plugin add hermoso@hermoso. Gemini CLI: gemini extensions install https://github.com/hermoso-ai/hermoso. Cursor, Cline and other skills-aware agents: npx skills add hermoso-ai/hermoso -a cursor, naming your own agent after -a. Each one installs five Hermoso skills that drive the hermoso CLI, and you sign in once with npx -y hermoso auth login. In Claude, ChatGPT, Perplexity, Le Chat or another chat app, add https://app.hermoso.ai/mcp as a custom connector and approve it with your Hermoso account. Meta Muse has no connector screen, so one chat message sets it up: see Hermoso in Muse. Full setup and the tool reference live on the MCP page.

Tools return durable asset URLs and structured results (never base64 blobs or prose the model has to re-parse), and long video renders submit to a job queue and poll to completion inside the tool call, so they work in every client. If you would rather click, the same engine runs as a web studio.

The command line, for agents that shell out

A terminal agent does not need the tool list loaded at all. The hermoso command answers the same questions on demand:

npx -y hermoso tools --search reddit   # every matching tool, name and one line
npx -y hermoso tools plan_ad           # one tool's full arguments
npx -y hermoso auth login              # sign in once, in a browser
npx -y hermoso call plan_ad --json '{"product":"…"}'   # run it

tools reads a registry bundled in the package, with no key, no network and no sign-in, so an agent can browse everything before anyone signs in. Only call spends, and only call needs the sign-in. On a machine with no browser, hermoso auth login --token hmk_… takes a key from MCP & CLI → Terminal & API keys. More on this path: ads from the command line.

Can an agent post on a schedule without you?

Yes. schedule_post puts a post in Hermoso’s own queue, with a caption per channel, so it goes out at its time whether or not the agent, the chat or your laptop is still running. list_scheduled reports each channel’s outcome separately, because “the job ran” and “the post went out” are different questions, and reschedule_post and cancel_scheduled move or pull a post that has not gone out. Ask an agent to plan a week in one message and the whole calendar lands in the queue, where you can also open it in the app.

Agents we have not found a way into yet

A page that says an agent works when it does not is worse than no page, so these have none. Each was checked against the vendor’s own documentation on 3 October 2026.

• OpenAI dots. Dots use supported ChatGPT plugins, and OpenAI has not documented whether a dot can use a developer-mode MCP app, which is how Hermoso connects to ChatGPT today. Hermoso works in ChatGPT and Codex.
• Cue. Manus’s new app of agents with their own responsibilities documents no custom MCP connector yet. Hermoso works in Manus.
• NanoClaw. It adds MCP servers per agent group from its own CLI, but it keeps credentials in a separate vault, and we have not verified how that vault hands Hermoso a key or an OAuth sign-in.
• The DeepSeek and Kimi chat apps. Neither documents a custom MCP option. Their agents do: DeepSeek Harness and Kimi Code.

Where to go next

See the platform-specific surfaces: marketing MCP, Meta Ads MCP, Google Ads MCP, marketing in Claude, Claude Code marketing and ads from the command line. For publishing specifically, start at the social media scheduler.

Frequently asked questions

What is an AI agent marketing tool?

It is software an AI assistant can call directly rather than software you click. Hermoso exposes ad research, creative generation, publishing, scheduling and paid campaign management as tools over the Model Context Protocol, so ChatGPT, Claude, Codex, Cursor, Cline or OpenClaw can carry out marketing work end to end instead of only advising you about it.

How do I connect Hermoso to Claude?

In Claude, open Customize, Connectors, add a custom connector with https://app.hermoso.ai/mcp and approve it with your Hermoso account. In a coding agent it is one command: /plugin marketplace add hermoso-ai/hermoso then /plugin install hermoso@hermoso in Claude Code, codex plugin marketplace add hermoso-ai/hermoso then codex plugin add hermoso@hermoso in Codex, gemini extensions install https://github.com/hermoso-ai/hermoso in Gemini CLI, or npx skills add hermoso-ai/hermoso -a followed by your agent’s name in Cursor and the rest. Then sign in once with npx -y hermoso auth login. Nothing is self-hosted and there are no API keys to manage.

Can an AI agent spend my ad budget without asking?

No. Every campaign on every paid platform is created paused, and the only tools that can arm real spend are the status switches, each of which refuses to run without an explicit confirmation flag. The same gate covers creating an enabled Google Ads ad group, ad or keyword under an already-enabled parent, because that would serve on the next auction.

Which AI clients does it work in?

Any MCP client. Setup guides cover Claude, ChatGPT, Claude Code, Cursor, Codex, Cline, OpenClaw, Grok Bot, Hermes, Gemini CLI, Perplexity, Le Chat, Claude Cowork, Perplexity Computer, Paperclip, Gemini, Grok Build, Kimi Code, DeepSeek Harness and Manus, plus Meta Muse. There is also an npm CLI for terminal agents that prefer to shell out, and installable skills that wrap the common workflows. The same account, credits and connections are shared across all of them and the web studio.

Does the agent see my social media passwords?

No. Every channel is connected through its own OAuth flow inside Hermoso's Connectors settings, in a browser, once. No password ever reaches the model. You then tick exactly which Pages, ad accounts and boards Hermoso may use, and the server re-verifies every id against the live connection, so an account you have not shared is unusable even by you.

What does it cost to use Hermoso from an agent?

API and MCP access are the product itself, and they are available on every plan including the free one. Hermoso charges credits, consumed by generation. Research and reads are cheap or free, and every model's exact per-render cost is published through a free tool the agent can call before it spends anything.

Which AI agents can post to social media with Hermoso?

Every agent on this page. Once Hermoso is connected, the agent can publish and schedule to Facebook, Instagram, Threads, TikTok, YouTube, X, LinkedIn, Pinterest, Bluesky, Telegram and Google Business Profile on your own accounts, and it can make the image or video first.

Can an AI agent post on a schedule without me?

Yes. schedule_post puts the post in Hermoso’s queue with a caption per channel, and it goes out at its time whether or not the agent is still running. list_scheduled reports what went out on each channel.

Does Hermoso work in OpenAI dots?

Not that we can document yet. Dots use supported ChatGPT plugins, and OpenAI has not said whether a dot can use a developer-mode MCP app, which is how Hermoso connects to ChatGPT. Hermoso works in ChatGPT and Codex today.

Should my agent use the MCP server or the CLI?

Whichever the agent can run. Chat apps can only add a URL, so they use the hosted MCP server. Agents with a shell do better with the CLI or the skills, because a tool list costs context in every turn and a command costs nothing until it runs. Both reach the same tools and the same account.

Related

An agent can also create and pay for its own Hermoso account, with no human at a sign-up form: see agents that sign up and pay.