Automate Cheerful With AI Agents
Cheerful has two machine-facing surfaces. The first is a remote MCP server: add it as a connector in Claude, ChatGPT, Cursor or any MCP client, sign in with your Cheerful account, and the assistant can read your campaigns, your replies, your creator lists and your credit balance — and search creators — on your behalf. The second is a command-line interface, so the same work scripts.
Both authenticate as you, and the difference between them matters. The MCP connector is read-only — it can see anything you can see and change nothing, so no assistant can send a message to a creator on your behalf. The CLI is not read-only. It drives the same API the web app does, which means it can create a campaign, launch one, send an email and spend credits. Treat a terminal session with the same care you would treat being logged in.
Cheerful runs a remote MCP server at mcp.cheerful.ai/mcp and publishes a command-line interface, so an AI assistant or a script can read your influencer campaigns, replies and creator lists using your own Cheerful account.
Three doors#
Connect Cheerful to Claude and ChatGPT
The MCP connector: one URL, an OAuth sign-in, and nine tools your assistant can call. Setup takes about a minute.
Every tool the connector exposes
All nine tools with their arguments, what comes back, what it costs — and the traps, which is the part nobody else publishes.
Run Cheerful from your terminal
The CLI: install with npm, sign in through the browser, 25 command groups, JSON output by default. Not read-only — it launches and sends.
Sign in to the Cheerful CLI
How authentication actually works, why there is no API key to paste, and what that means for unattended scripts.
What an agent can actually do with this#
The interesting half is not creator search. Every influencer platform has search, and a lookup tool bolted onto an assistant is still a lookup tool. What the Cheerful connector exposes that no comparable platform does is the operational half of the job: the campaigns you are running, the mailboxes they send from, the replies that came back, the creators who opted in, and the credits it all consumed.
That changes the kind of question you can ask. "Find me fitness creators" is a search. "Which of my campaigns has the worst reply rate this month, and which mailbox is it sending from?" is a question about work in progress, and it needs four different systems to answer. Through the connector it is one sentence.
| You ask | It reads |
|---|---|
| Which of my campaigns has the worst reply rate this month? | campaign_report |
| Who replied to me yesterday and what did they say? | inbox_read |
| Draft me a summary of every creator still waiting on an answer. | inbox_read |
| Which mailbox is the spring gifting campaign actually sending from? | campaign_read |
| How many creators opted in across all my campaigns this quarter? | campaign_report |
| Am I connected to Shopify, and is the connection healthy? | integration_status |
Read-only, and why that is the right first step#
Every tool on the connector is annotated read-only, and that annotation is what a client uses to decide whether a call needs your approval. There is no write tool to mis-annotate, no catch-all tool that accepts both safe and unsafe operations, and no path by which an assistant can send a message to a creator on your behalf.
That is a deliberate first phase rather than a limitation we are shy about. Outreach that goes out under your name is the part of this product that most deserves a human at the keyboard, and an assistant that can read everything and change nothing is the version of this worth shipping first. Write tools are designed and not shipped; when they land they will be separate, individually annotated, and obvious.
What it costs#
Reads are free. Creator search and creator profiles consume credits from the same balance the web app uses, at the same rate, because they are the same searches. Nothing on this surface is billed separately for being called by an agent rather than by a human, and there is no developer plan to buy.
If a call comes back with a 402, that is your credit balance rather than a fault. billing_usage reports what is left and when it renews.
Guides for running the work itself#
The pages above are about wiring Cheerful to something else. These are about the job: how outreach, replies and gifting actually run, written for the person doing them rather than the person integrating them. Each one states what the product does not do, which is usually the part worth reading first.
Manage your campaigns from Claude or ChatGPT
The non-technical half of the connector: what an assistant can see, what it cannot do, and where it beats a dashboard.
Run outreach from your own Gmail
Sending from your own mailbox rather than a platform relay, a Google Sheet as the roster, and the limits that apply.
Reply to influencer emails automatically
What gets read before a reply is drafted, where approval sits, and when to take the thread back.
Track gifting from shipment to ROI
Per-creator codes, draft orders, post detection — and the honest limits of attributing gifting revenue.
FTC disclosure rules
What counts as a material connection, what "clear and conspicuous" means per format, and where the brand's liability sits.
What this does not do#
- The MCP connector is read-only. No tool on it sends an email, launches a campaign, edits a creator list or places a gifting order.
- The CLI is not read-only, and that is worth saying plainly: it can create and launch campaigns, send email, edit drafts and spend credits, because it calls the same API the web app calls.
- Neither surface reaches more than your own account does — your own work plus whatever your teams share. Neither can reach another customer’s data, or data your own login is not entitled to.
- Creator search takes 90–155 seconds and a cold creator profile around 60 seconds. Some clients will time out before we answer.
- There is no API key to paste. Both surfaces authenticate through a browser sign-in, which means unattended use — CI, cron, a shared runner with no browser — is not supported today.
- The CLI is published for macOS (Apple Silicon and Intel), Linux x64 and Windows x64. There is no Linux arm64 build.
Frequently asked questions#
- Do I need to be a developer to use this?
- For the MCP connector, no. It is a URL you paste into Claude’s or ChatGPT’s connector settings and a sign-in you complete in your browser. The CLI is a developer tool and assumes a terminal.
- Does it cost extra?
- No. There is no developer plan and no per-call charge. Reads are free; creator searches consume the same credits the web app does.
- Can it send emails on my behalf?
- No. Every tool is read-only. Sending happens in the Cheerful web app, where messages can be held for one-click approval.
- Which assistants work with it?
- Any client that speaks the Model Context Protocol over streamable HTTP with OAuth. That includes Claude, ChatGPT, Cursor, VS Code and Windsurf.
- Can my whole team use it?
- Yes. Each person connects with their own Cheerful account and sees exactly what that account is entitled to see, including campaigns their teams share with them.