MCP · Paperclip
Marketing MCP for Paperclip
Paperclip is where you manage the agents doing the work. Connect Spreeflo and marketing becomes one of the functions they can actually run.
Agent orchestration·MIT licensed·Adapters for Claude Code, Codex, Cursor
{
"mcpServers": {
"spreeflo": {
"url": "https://mcp.spreeflo.com/mcp",
"headers": { "Authorization": "Bearer <YOUR_API_KEY>" }
}
}
}Writes apply to your live workspace immediately — there's no approval queue here. Keep a human in the loop and check Paperclip's confirmation settings first.
The agent
What is Paperclip?
Paperclip is an open-source app for managing AI agents at work — agents, projects, tasks, approvals and costs in one place. It ships adapters for Claude Code, Codex, Gemini CLI, Cursor and others, so the agent that reaches Spreeflo is whichever one you assign.
Many agents, one place
Assign marketing work to an agent the way you would assign a ticket, and see what it cost.
Adapter-based
Claude Code, Codex, Cursor and others plug in — so Spreeflo is configured on the adapter that runs the task.
Approvals built in
A useful pairing with Spreeflo’s MCP surface, where writes apply immediately and the review step lives outside Spreeflo.
Setup
Connect Paperclip in about a minute
A Spreeflo API key, and one entry in Paperclip.
Create an API key
In the Spreeflo app, go to Settings → API Keys and click + Add API Key. Name it Paperclip so you can revoke this one connection later without touching the others. It goes in the Authorization header — never in the server URL.
Add the Spreeflo MCP server
Adapter config- 1.Pick the adapter that will run marketing work — Claude Code and Codex are the common choices.
- 2.Add Spreeflo as an MCP server on that adapter using the form below.
- 3.Assign the agent a task and check its approval settings before letting it write.
{
"mcpServers": {
"spreeflo": {
"url": "https://mcp.spreeflo.com/mcp",
"headers": { "Authorization": "Bearer <YOUR_API_KEY>" }
}
}
}Paperclip also publishes its own MCP server for managing Paperclip itself — that is the opposite direction to what this page describes.
Check it works
Ask Paperclip for something small — "list my Spreeflo audiences". A working connection is one where the assigned agent returns real audience names in its task output.
Full setup guide and troubleshootingIn practice
What Paperclip does with it
Two ordinary requests, and the Spreeflo tools behind them.
Task: prepare the Q3 nurture sequence. Assign it to the marketing agent.
The assigned agent read the format spec and built the sequence as a draft journey:
read_format_doccreate_journey_sequenceget_sequence
Created as a draft — there is no MCP tool to start a journey.
What did the marketing agent actually produce this sprint?
Paperclip tracks the agent’s cost and output; Spreeflo shows what landed:
list_campaignslist_email_templateslist_segments
The audit trail on the Spreeflo side names the human whose key authorised each change.
Running it
Where the connection actually lives
Paperclip manages agents rather than being one, so nothing here connects to Spreeflo directly. The configuration goes on whichever adapter runs the task — and once you see that, the rest of the setup follows.
Configure the adapter, not Paperclip
Spreeflo is added to Claude Code, Codex, Cursor or whichever adapter you assign the work to, using that tool’s own MCP configuration. Looking for an MCP setting in Paperclip itself is the usual first wrong turn — it manages the agent that holds the connection, not the connection.
Approvals are the review step Spreeflo lacks
Spreeflo’s MCP surface applies writes immediately; there is no proposal queue. Paperclip’s approval gate sits upstream of the tool call, which makes this pairing unusually well matched — it supplies precisely the control the protocol side does not have.
Two different records, both useful
Paperclip tracks what an agent cost and what task it was working. Spreeflo records what changed and whose key authorised it. Neither is complete alone; together they answer “what did we spend, and what did it actually do to the workspace”.
Paperclip’s own MCP server points the other way
Paperclip publishes a server for managing Paperclip — tasks, agents, projects — from an outside client. That is the reverse of this page, which is about agents inside Paperclip reaching your marketing workspace. Easy to conflate when reading its docs.
Every agent you assign shares the key you configured
The credential belongs to the adapter, so a fleet of agents running through the same adapter all act as the same Spreeflo user. Cost and task attribution stay separate in Paperclip, but the Spreeflo audit trail will name one person for all of it. If you need changes traceable to individual operators, give each their own adapter configuration and their own key.
Why this pairing
Paperclip + Spreeflo
- Marketing becomes an assignable, costed function
- Works through whichever adapter you already trust
- Approvals sit where Spreeflo’s MCP surface has none
- Spreeflo’s audit trail attributes changes to a real person
- No MCP tool for irreversible actions
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.
What an assigned agent touches depends on the task you wrote — sequence and template work for campaign briefs, the reporting tools when you ask what a sprint produced.
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.
Whichever one you assign marketing work to — Claude Code and Codex are the common picks, and both read a standard MCP server block. If you route different marketing tasks through different adapters, each needs its own entry; there is no shared connection at the Paperclip level for them to inherit.
They gate the agent before it acts, which is upstream of the tool call and therefore effective. It is worth being precise about the mechanism: Spreeflo itself does not hold a write pending review, so the protection is entirely Paperclip’s. Configure the approval settings deliberately rather than assuming a queue exists on our side.
It depends on what you need to prove afterwards. One key per adapter is simpler and usually enough. One key per operator makes Spreeflo’s audit trail attribute changes to the right person, which matters if several people direct agents against the same workspace.
No. Spreeflo runs a hosted MCP server, so Paperclip calls it directly over Streamable HTTP — nothing to install, nothing to self-host, no SDK to keep up to date. A server URL and a sign-in (or, for a headless agent, an API key in the Authorization header) are the whole integration.
It cannot delete anything or start a send. Spreeflo has no MCP tool for deletes, starting or resuming a campaign or journey, sending or scheduling a broadcast, or publishing a form, rather than leaving them to client settings. Those stay in the Spreeflo 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.
The key resolves to exactly one workspace, and every action runs as the person who owns that key — their membership re-verified on each request. The audit trail names a real person, so agent-made changes are attributable.