Spreeflo MCP server

MCP (Model Context Protocol) is an open standard for connecting AI applications to external systems. Spreeflo runs a hosted MCP server, so your own AI client — Claude, Claude Code, Cursor, and other clients that support remote MCP servers — can read and build inside your Spreeflo workspace using the same tools that power Spreeflo's own AI.

You connect it once with a Spreeflo API key. After that, your client can look up campaign performance, inspect segments, draft emails, build journeys, and read form responses — without you copying anything between tabs.

What you can do with it

Once connected, you can ask your AI client things like:

  • "What did our last three campaigns do on open and click rate?"
  • "Build me a 3-email welcome journey for the Newsletter audience and leave it in draft."
  • "Create a segment of contacts who visited /pricing twice in the last 14 days but never opened an email."
  • "Summarise the responses to our onboarding survey."
  • "Draft a web push template for the Black Friday campaign."
  • "Pause the campaign that's going out to the trial audience."

    • See the full tool catalog for everything the server exposes.

    How it works

    The server is a remote MCP server served over Streamable HTTP. Your client posts to a single endpoint and authenticates with a Spreeflo API key — the same key you'd use for the Spreeflo REST API. 

    POST https://mcp.spreeflo.com/mcp
Authorization: Bearer <YOUR_API_KEY>

    There is no OAuth flow, no connector directory to browse, and no separate MCP credential — a bearer API key on that endpoint is the whole connection. The server is stateless: every request builds a fresh server, so there are no sessions to keep alive and no sticky routing to worry about.

    The safety model

    Four rules govern everything an external agent can do in your workspace:

  • Writes apply immediately. There is no approval queue on this surface. When your client calls a write tool, the change lands in the live workspace at that moment.

  • Your client controls confirmation. Spreeflo will not prompt you before a write. Every tool is annotated so a well-behaved client knows which ones modify data, but it's the client that decides whether to ask you first.

  • Irreversible and outward-facing actions are refused. Every delete, starting or resuming a campaign or journey, and publishing a form are refused server-side — on every client, regardless of its settings. Those are only ever done by a person signed in to the Spreeflo app.

  • Everything is audited. Each tool call is its own agent run with an audit record, and each applied write is recorded as a change attributed to the API key's owner. You can see exactly what an external agent did in your workspace's change history.
    • Warning: Writes over MCP apply immediately to your live workspace. Before you start, check how your AI client handles write tools — Spreeflo will not ask you to confirm.

    There's a deliberate asymmetry worth knowing: an agent can pause a campaign but never start one. Every mistake that's reachable over MCP fails toward sending less, not more.

    Cost and limits

    Your connected client owns all the model reasoning; Spreeflo only exposes deterministic tools. The model cost sits with that client, not with Spreeflo.

    The only governor is a rolling rate limit of about 120 tool calls per minute per workspace. There's no daily quota — the window drains continuously, so if you hit it, waiting a few seconds is enough to carry on.

    Access and scope

    A connection is scoped to one team and one person. The API key resolves to exactly one workspace, and every action runs as the human who owns that key — with their team membership re-verified on every request. Remove someone from the team and their keys stop working immediately.

    To work in a second team, add a second server entry in your client with a key belonging to that team. Full details are in Connect your AI client.

    Spreeflo has two MCP servers

    These docs cover the hosted server. There is also a local server that ships with the Spreeflo CLI, and the two behave very differently.

    Hosted MCP serverspreeflo mcp (CLI)
    TransportRemote HTTP, bearer API keyLocal stdio, runs as you
    SurfaceAround 67 curated toolsThe whole REST API through two tools (get_schema, call_api) plus read_format_doc
    Irreversible actionsRefused server-sideAllowed — deletes, campaign start/resume, form publish
    SafeguardDangerous actions are simply unavailableOnly your client's confirmation policy
    ForAnyone connecting an AI client to their workspaceDevelopers who could equally run curl

    Next steps

  • Connect your AI client — create a key, add the server, verify the connection.
  • MCP tools — the full catalog, and what is refused.
  • MCP vs. the in-app AI — same engine, different surface.
    • Need Immediate Help? Contact our support team at support@spreeflo.com or check our other guides for more information.