For AI agents
Give your agent a marketing stack
Spreeflo exposes audiences, campaigns, journeys, templates and forms as MCP tools. No CLI to install, no SDK to depend on — your agent calls a hosted server directly.
Writes apply to your live workspace immediately — there's no approval queue on this surface. Keep a human in the loop, and check your agent's confirmation settings before letting it run unattended.
Supported agents
Connect once, from anywhere
Each guide covers how that agent registers an MCP server, and what it does with Spreeflo once it has one.
OpenClaw
Open source
OpenClaw skills are MCP servers, so Spreeflo is not an integration to build — it is a skill you enable. Then your agent can run marketing from any chat app it listens on.
Skill (MCP server)Claude Cowork
Anthropic
Give Cowork your Spreeflo workspace and it can build segments, draft campaigns, and pull performance while it works alongside your other tools.
Connector dialogCodex
OpenAI
Codex already lives in your repo and your terminal. Point it at Spreeflo and it can ship the marketing side of a release without leaving them.
config.tomlPerplexity Computer
Perplexity
Computer researches the web and acts on what it finds. Connect Spreeflo and the research can land as a segment, a draft, or a report in your workspace.
Custom connectorHermes Agent
Nous Research
Hermes runs on your own infrastructure and remembers what it learns. Connect Spreeflo and your marketing work becomes something it gets better at.
MCP server configNanoClaw
nanoco.ai
NanoClaw keeps every agent in its own container. Connect Spreeflo and your marketing tools are reachable without handing the agent your whole machine.
Agent templatePaperclip
Paperclip
Paperclip is where you manage the agents doing the work. Connect Spreeflo and marketing becomes one of the functions they can actually run.
Adapter configUsing something else? Any client that speaks remote MCP over Streamable HTTP works — see the MCP server .
Why connect it
Your marketing stack, in the tool you're already thinking in
No copy-pasting CSVs into a chat window, no screenshots of dashboards. Your assistant reads the live workspace and does the work in it.
Ask in plain language
Describe the segment, the campaign, or the report you want. Your assistant translates it into the right tool calls and shows you what it did.
Reads your real data
Live audiences, contacts, tags, and performance numbers — answers come from the workspace itself, not from a stale export you pasted in.
Builds, not just advises
Segments, journeys, email and web push templates, and forms are all authorable over MCP. The output is a real draft in your workspace.
One URL, one sign-in
Add the hosted endpoint, sign in to Spreeflo and approve. Nothing to install, nothing to self-host, and no key to paste into your client.
Dangerous actions removed
Deletes, campaign and broadcast sends, and form publishes have no tool on the server — not left to whether your client remembered to ask you first.
Your client does the thinking
Spreeflo exposes deterministic tools and nothing more. The model reasoning — and whatever it costs — stays with the client you already pay for.
Coverage
Everything it can touch
The same surfaces you work with in the app, exposed as tools — and a short list that is deliberately off limits.
Audience
- Audiences and contacts
- Segments, static and dynamic
- Tags
- Custom contact attributes
- CSV import
Sending
- Campaigns and journeys
- Email templates
- Web push templates
- Sender and domain identities
Capture
- Forms and question types
- Inline and popup embeds
- Response summaries
- CSV export of responses
Insight
- Campaign and web push performance
- Tracked events and pages
- Link clicks
- Form analytics
What MCP can’t do
These are irreversible or they reach real people, so they are removed from the surface rather than left to your client's confirmation settings. A person does them in the app.
Supercharge your marketing automation
FAQ
Questions, answered.
The Model Context Protocol is an open standard for connecting AI assistants to real tools and data. Instead of pasting exports into a chat window, your assistant calls a server directly — reading your live workspace and acting in it. Spreeflo runs a hosted MCP server, so there is nothing to install or self-host.
Any client that speaks remote MCP over Streamable HTTP — ChatGPT, Claude and Claude Cowork, Claude Code, Cursor, and others. They connect by signing in: you approve the connection from your Spreeflo account and pick the workspace. Headless agents that cannot open a browser send a Spreeflo API key in the Authorization header instead. Either way they reach the same tools.
Audiences and contacts, segments and tags, campaigns and journeys, email and web push templates, forms, and reporting. Your assistant can read all of it and create or edit most of it — the same capabilities you have in the app, on the same plan gates.
It cannot delete anything or start a send. The riskiest actions are removed from the surface rather than left to your client settings: there is no tool for deletes, starting or resuming a campaign or journey, sending or scheduling a broadcast, or publishing a form. Those stay in the app, where a person does them. Journeys you have already switched on keep running, though, so editing a contact, its tags, a segment or an email template, or importing contacts, can set one off, just as the same edit would in the app.
Not on this surface — a write applies to your live workspace the moment your assistant calls it. Whether you get asked first is entirely your client’s decision, so check its confirmation settings before you start. If you want a review step, use Spreeflo’s in-app AI, which queues changes as proposals instead.
Yes. When you approve the connection, leave Make changes unticked. Your assistant can still read everything, but it is offered no tool that changes your workspace. To allow changes later, reconnect and tick it.
Your connected client does all the model reasoning, so the model cost sits with that client rather than with Spreeflo. On our side the only governor is a rolling limit of roughly 120 tool calls per minute per workspace — there is no daily quota to wait out, and the window drains continuously.